author
int64 658
755k
| date
stringlengths 19
19
| timezone
int64 -46,800
43.2k
| hash
stringlengths 40
40
| message
stringlengths 5
490
| mods
list | language
stringclasses 20
values | license
stringclasses 3
values | repo
stringlengths 5
68
| original_message
stringlengths 12
491
|
|---|---|---|---|---|---|---|---|---|---|
499,333 |
13.05.2022 20:57:53
| -28,800 |
67742521f191b423f277f004a71c2d2b41545019
|
fix evaldataset
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -376,7 +376,8 @@ class Trainer(object):\nassert self.mode == 'train', \"Model not in 'train' mode\"\nInit_mark = False\nif validate:\n- self.cfg.EvalDataset = create(\"EvalDataset\")()\n+ self.cfg['EvalDataset'] = self.cfg.EvalDataset = create(\n+ \"EvalDataset\")()\nsync_bn = (getattr(self.cfg, 'norm_type', None) == 'sync_bn' and\nself.cfg.use_gpu and self._nranks > 1)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix evaldataset (#5978)
|
499,304 |
13.05.2022 20:58:22
| -28,800 |
a972c39e26a40c90c2ef4c1c4a06f8e9fba15509
|
support fuse conv bn when export model
|
[
{
"change_type": "MODIFY",
"old_path": "configs/runtime.yml",
"new_path": "configs/runtime.yml",
"diff": "@@ -10,3 +10,4 @@ export:\npost_process: True # Whether post-processing is included in the network when export model.\nnms: True # Whether NMS is included in the network when export model.\nbenchmark: False # It is used to testing model performance, if set `True`, post-process and NMS will not be exported.\n+ fuse_conv_bn: False\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -44,6 +44,7 @@ from ppdet.metrics import RBoxMetric, JDEDetMetric, SNIPERCOCOMetric\nfrom ppdet.data.source.sniper_coco import SniperCOCODataSet\nfrom ppdet.data.source.category import get_categories\nimport ppdet.utils.stats as stats\n+from ppdet.utils.fuse_utils import fuse_conv_bn\nfrom ppdet.utils import profiler\nfrom .callbacks import Callback, ComposeCallback, LogPrinter, Checkpointer, WiferFaceEval, VisualDLWriter, SniperProposalsGenerator, WandbCallback\n@@ -770,6 +771,11 @@ class Trainer(object):\ndef export(self, output_dir='output_inference'):\nself.model.eval()\n+\n+ if hasattr(self.cfg, 'export') and 'fuse_conv_bn' in self.cfg[\n+ 'export'] and self.cfg['export']['fuse_conv_bn']:\n+ self.model = fuse_conv_bn(self.model)\n+\nmodel_name = os.path.splitext(os.path.split(self.cfg.filename)[-1])[0]\nsave_dir = os.path.join(output_dir, model_name)\nif not os.path.exists(save_dir):\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "ppdet/utils/fuse_utils.py",
"diff": "+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing, software\n+# distributed under the License is distributed on an \"AS IS\" BASIS,\n+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+# See the License for the specific language governing permissions and\n+# limitations under the License.\n+\n+import copy\n+import paddle\n+import paddle.nn as nn\n+\n+__all__ = ['fuse_conv_bn']\n+\n+\n+def fuse_conv_bn(model):\n+ is_train = False\n+ if model.training:\n+ model.eval()\n+ is_train = True\n+ fuse_list = []\n+ tmp_pair = [None, None]\n+ for name, layer in model.named_sublayers():\n+ if isinstance(layer, nn.Conv2D):\n+ tmp_pair[0] = name\n+ if isinstance(layer, nn.BatchNorm2D):\n+ tmp_pair[1] = name\n+\n+ if tmp_pair[0] and tmp_pair[1] and len(tmp_pair) == 2:\n+ fuse_list.append(tmp_pair)\n+ tmp_pair = [None, None]\n+ model = fuse_layers(model, fuse_list)\n+ if is_train:\n+ model.train()\n+ return model\n+\n+\n+def find_parent_layer_and_sub_name(model, name):\n+ \"\"\"\n+ Given the model and the name of a layer, find the parent layer and\n+ the sub_name of the layer.\n+ For example, if name is 'block_1/convbn_1/conv_1', the parent layer is\n+ 'block_1/convbn_1' and the sub_name is `conv_1`.\n+ Args:\n+ model(paddle.nn.Layer): the model to be quantized.\n+ name(string): the name of a layer\n+\n+ Returns:\n+ parent_layer, subname\n+ \"\"\"\n+ assert isinstance(model, nn.Layer), \\\n+ \"The model must be the instance of paddle.nn.Layer.\"\n+ assert len(name) > 0, \"The input (name) should not be empty.\"\n+\n+ last_idx = 0\n+ idx = 0\n+ parent_layer = model\n+ while idx < len(name):\n+ if name[idx] == '.':\n+ sub_name = name[last_idx:idx]\n+ if hasattr(parent_layer, sub_name):\n+ parent_layer = getattr(parent_layer, sub_name)\n+ last_idx = idx + 1\n+ idx += 1\n+ sub_name = name[last_idx:idx]\n+ return parent_layer, sub_name\n+\n+\n+class Identity(nn.Layer):\n+ '''a layer to replace bn or relu layers'''\n+\n+ def __init__(self, *args, **kwargs):\n+ super(Identity, self).__init__()\n+\n+ def forward(self, input):\n+ return input\n+\n+\n+def fuse_layers(model, layers_to_fuse, inplace=False):\n+ '''\n+ fuse layers in layers_to_fuse\n+\n+ Args:\n+ model(nn.Layer): The model to be fused.\n+ layers_to_fuse(list): The layers' names to be fused. For\n+ example,\"fuse_list = [[\"conv1\", \"bn1\"], [\"conv2\", \"bn2\"]]\".\n+ A TypeError would be raised if \"fuse\" was set as\n+ True but \"fuse_list\" was None.\n+ Default: None.\n+ inplace(bool): Whether apply fusing to the input model.\n+ Default: False.\n+\n+ Return\n+ fused_model(paddle.nn.Layer): The fused model.\n+ '''\n+ if not inplace:\n+ model = copy.deepcopy(model)\n+ for layers_list in layers_to_fuse:\n+ layer_list = []\n+ for layer_name in layers_list:\n+ parent_layer, sub_name = find_parent_layer_and_sub_name(model,\n+ layer_name)\n+ layer_list.append(getattr(parent_layer, sub_name))\n+ new_layers = _fuse_func(layer_list)\n+ for i, item in enumerate(layers_list):\n+ parent_layer, sub_name = find_parent_layer_and_sub_name(model, item)\n+ setattr(parent_layer, sub_name, new_layers[i])\n+ return model\n+\n+\n+def _fuse_func(layer_list):\n+ '''choose the fuser method and fuse layers'''\n+ types = tuple(type(m) for m in layer_list)\n+ fusion_method = types_to_fusion_method.get(types, None)\n+ new_layers = [None] * len(layer_list)\n+ fused_layer = fusion_method(*layer_list)\n+ for handle_id, pre_hook_fn in layer_list[0]._forward_pre_hooks.items():\n+ fused_layer.register_forward_pre_hook(pre_hook_fn)\n+ del layer_list[0]._forward_pre_hooks[handle_id]\n+ for handle_id, hook_fn in layer_list[-1]._forward_post_hooks.items():\n+ fused_layer.register_forward_post_hook(hook_fn)\n+ del layer_list[-1]._forward_post_hooks[handle_id]\n+ new_layers[0] = fused_layer\n+ for i in range(1, len(layer_list)):\n+ identity = Identity()\n+ identity.training = layer_list[0].training\n+ new_layers[i] = identity\n+ return new_layers\n+\n+\n+def _fuse_conv_bn(conv, bn):\n+ '''fuse conv and bn for train or eval'''\n+ assert(conv.training == bn.training),\\\n+ \"Conv and BN both must be in the same mode (train or eval).\"\n+ if conv.training:\n+ assert bn._num_features == conv._out_channels, 'Output channel of Conv2d must match num_features of BatchNorm2d'\n+ raise NotImplementedError\n+ else:\n+ return _fuse_conv_bn_eval(conv, bn)\n+\n+\n+def _fuse_conv_bn_eval(conv, bn):\n+ '''fuse conv and bn for eval'''\n+ assert (not (conv.training or bn.training)), \"Fusion only for eval!\"\n+ fused_conv = copy.deepcopy(conv)\n+\n+ fused_weight, fused_bias = _fuse_conv_bn_weights(\n+ fused_conv.weight, fused_conv.bias, bn._mean, bn._variance, bn._epsilon,\n+ bn.weight, bn.bias)\n+ fused_conv.weight.set_value(fused_weight)\n+ if fused_conv.bias is None:\n+ fused_conv.bias = paddle.create_parameter(\n+ shape=[fused_conv._out_channels], is_bias=True, dtype=bn.bias.dtype)\n+ fused_conv.bias.set_value(fused_bias)\n+ return fused_conv\n+\n+\n+def _fuse_conv_bn_weights(conv_w, conv_b, bn_rm, bn_rv, bn_eps, bn_w, bn_b):\n+ '''fuse weights and bias of conv and bn'''\n+ if conv_b is None:\n+ conv_b = paddle.zeros_like(bn_rm)\n+ if bn_w is None:\n+ bn_w = paddle.ones_like(bn_rm)\n+ if bn_b is None:\n+ bn_b = paddle.zeros_like(bn_rm)\n+ bn_var_rsqrt = paddle.rsqrt(bn_rv + bn_eps)\n+ conv_w = conv_w * \\\n+ (bn_w * bn_var_rsqrt).reshape([-1] + [1] * (len(conv_w.shape) - 1))\n+ conv_b = (conv_b - bn_rm) * bn_var_rsqrt * bn_w + bn_b\n+ return conv_w, conv_b\n+\n+\n+types_to_fusion_method = {(nn.Conv2D, nn.BatchNorm2D): _fuse_conv_bn, }\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
support fuse conv bn when export model (#5977)
|
499,301 |
23.05.2022 11:15:25
| -28,800 |
ea5f339ac6178360cea1ed0c30fa72dd40cc3802
|
reorg optimizer
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "ppdet/optimizer/__init__.py",
"diff": "+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing, software\n+# distributed under the License is distributed on an \"AS IS\" BASIS,\n+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+# See the License for the specific language governing permissions and\n+# limitations under the License.\n+\n+from .optimizer import *\n+from .ema import ModelEMA\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "ppdet/optimizer/ema.py",
"diff": "+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing, software\n+# distributed under the License is distributed on an \"AS IS\" BASIS,\n+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+# See the License for the specific language governing permissions and\n+# limitations under the License.\n+\n+from __future__ import absolute_import\n+from __future__ import division\n+from __future__ import print_function\n+\n+import math\n+import paddle\n+import weakref\n+\n+\n+class ModelEMA(object):\n+ \"\"\"\n+ Exponential Weighted Average for Deep Neutal Networks\n+ Args:\n+ model (nn.Layer): Detector of model.\n+ decay (int): The decay used for updating ema parameter.\n+ Ema's parameter are updated with the formula:\n+ `ema_param = decay * ema_param + (1 - decay) * cur_param`.\n+ Defaults is 0.9998.\n+ ema_decay_type (str): type in ['threshold', 'normal', 'exponential'],\n+ 'threshold' as default.\n+ cycle_epoch (int): The epoch of interval to reset ema_param and\n+ step. Defaults is -1, which means not reset. Its function is to\n+ add a regular effect to ema, which is set according to experience\n+ and is effective when the total training epoch is large.\n+ \"\"\"\n+\n+ def __init__(self,\n+ model,\n+ decay=0.9998,\n+ ema_decay_type='threshold',\n+ cycle_epoch=-1):\n+ self.step = 0\n+ self.epoch = 0\n+ self.decay = decay\n+ self.state_dict = dict()\n+ for k, v in model.state_dict().items():\n+ self.state_dict[k] = paddle.zeros_like(v)\n+ self.ema_decay_type = ema_decay_type\n+ self.cycle_epoch = cycle_epoch\n+\n+ self._model_state = {\n+ k: weakref.ref(p)\n+ for k, p in model.state_dict().items()\n+ }\n+\n+ def reset(self):\n+ self.step = 0\n+ self.epoch = 0\n+ for k, v in self.state_dict.items():\n+ self.state_dict[k] = paddle.zeros_like(v)\n+\n+ def resume(self, state_dict, step=0):\n+ for k, v in state_dict.items():\n+ if k in self.state_dict:\n+ self.state_dict[k] = v\n+ self.step = step\n+\n+ def update(self, model=None):\n+ if self.ema_decay_type == 'threshold':\n+ decay = min(self.decay, (1 + self.step) / (10 + self.step))\n+ elif self.ema_decay_type == 'exponential':\n+ decay = self.decay * (1 - math.exp(-(self.step + 1) / 2000))\n+ else:\n+ decay = self.decay\n+ self._decay = decay\n+\n+ if model is not None:\n+ model_dict = model.state_dict()\n+ else:\n+ model_dict = {k: p() for k, p in self._model_state.items()}\n+ assert all(\n+ [v is not None for _, v in model_dict.items()]), 'python gc.'\n+\n+ for k, v in self.state_dict.items():\n+ v = decay * v + (1 - decay) * model_dict[k]\n+ v.stop_gradient = True\n+ self.state_dict[k] = v\n+ self.step += 1\n+\n+ def apply(self):\n+ if self.step == 0:\n+ return self.state_dict\n+ state_dict = dict()\n+ for k, v in self.state_dict.items():\n+ if self.ema_decay_type != 'exponential':\n+ v = v / (1 - self._decay**self.step)\n+ v.stop_gradient = True\n+ state_dict[k] = v\n+ self.epoch += 1\n+ if self.cycle_epoch > 0 and self.epoch == self.cycle_epoch:\n+ self.reset()\n+\n+ return state_dict\n"
},
{
"change_type": "RENAME",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer/optimizer.py",
"diff": "@@ -18,7 +18,6 @@ from __future__ import print_function\nimport sys\nimport math\n-import weakref\nimport paddle\nimport paddle.nn as nn\n@@ -360,89 +359,3 @@ class OptimizerBuilder():\nparameters=params,\ngrad_clip=grad_clip,\n**optim_args)\n-\n-\n-class ModelEMA(object):\n- \"\"\"\n- Exponential Weighted Average for Deep Neutal Networks\n- Args:\n- model (nn.Layer): Detector of model.\n- decay (int): The decay used for updating ema parameter.\n- Ema's parameter are updated with the formula:\n- `ema_param = decay * ema_param + (1 - decay) * cur_param`.\n- Defaults is 0.9998.\n- ema_decay_type (str): type in ['threshold', 'normal', 'exponential'],\n- 'threshold' as default.\n- cycle_epoch (int): The epoch of interval to reset ema_param and\n- step. Defaults is -1, which means not reset. Its function is to\n- add a regular effect to ema, which is set according to experience\n- and is effective when the total training epoch is large.\n- \"\"\"\n-\n- def __init__(self,\n- model,\n- decay=0.9998,\n- ema_decay_type='threshold',\n- cycle_epoch=-1):\n- self.step = 0\n- self.epoch = 0\n- self.decay = decay\n- self.state_dict = dict()\n- for k, v in model.state_dict().items():\n- self.state_dict[k] = paddle.zeros_like(v)\n- self.ema_decay_type = ema_decay_type\n- self.cycle_epoch = cycle_epoch\n-\n- self._model_state = {\n- k: weakref.ref(p)\n- for k, p in model.state_dict().items()\n- }\n-\n- def reset(self):\n- self.step = 0\n- self.epoch = 0\n- for k, v in self.state_dict.items():\n- self.state_dict[k] = paddle.zeros_like(v)\n-\n- def resume(self, state_dict, step=0):\n- for k, v in state_dict.items():\n- if k in self.state_dict:\n- self.state_dict[k] = v\n- self.step = step\n-\n- def update(self, model=None):\n- if self.ema_decay_type == 'threshold':\n- decay = min(self.decay, (1 + self.step) / (10 + self.step))\n- elif self.ema_decay_type == 'exponential':\n- decay = self.decay * (1 - math.exp(-(self.step + 1) / 2000))\n- else:\n- decay = self.decay\n- self._decay = decay\n-\n- if model is not None:\n- model_dict = model.state_dict()\n- else:\n- model_dict = {k: p() for k, p in self._model_state.items()}\n- assert all(\n- [v is not None for _, v in model_dict.items()]), 'python gc.'\n-\n- for k, v in self.state_dict.items():\n- v = decay * v + (1 - decay) * model_dict[k]\n- v.stop_gradient = True\n- self.state_dict[k] = v\n- self.step += 1\n-\n- def apply(self):\n- if self.step == 0:\n- return self.state_dict\n- state_dict = dict()\n- for k, v in self.state_dict.items():\n- if self.ema_decay_type != 'exponential':\n- v = v / (1 - self._decay**self.step)\n- v.stop_gradient = True\n- state_dict[k] = v\n- self.epoch += 1\n- if self.cycle_epoch > 0 and self.epoch == self.cycle_epoch:\n- self.reset()\n-\n- return state_dict\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
reorg optimizer (#6016)
|
499,299 |
23.05.2022 12:22:17
| -28,800 |
b185733495f0600c8e178e331f04f98d5d4ae25b
|
fix solov2 not support multi-images
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -224,7 +224,7 @@ class Detector(object):\nfor k, v in res.items():\nresults[k].append(v)\nfor k, v in results.items():\n- if k != 'masks':\n+ if k not in ['masks', 'segm']:\nresults[k] = np.concatenate(v)\nreturn results\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix solov2 not support multi-images (#6019)
|
499,333 |
24.05.2022 10:36:08
| -28,800 |
1d8c3a7edffc6c899f3b7ab3eb95b81ef300be0b
|
enhance shm utils
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/shm_utils.py",
"new_path": "ppdet/data/shm_utils.py",
"diff": "@@ -34,6 +34,9 @@ SHM_DEFAULT_MOUNT = '/dev/shm'\ndef _parse_size_in_M(size_str):\n+ if size_str[-1] == 'B':\n+ num, unit = size_str[:-2], size_str[-2]\n+ else:\nnum, unit = size_str[:-1], size_str[-1]\nassert unit in SIZE_UNIT, \\\n\"unknown shm size unit {}\".format(unit)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
enhance shm utils (#6042)
|
499,299 |
25.05.2022 10:49:48
| -28,800 |
84faecbca8748f1da572d0da12e3c0214f3d84bf
|
add cpp infer support for solov2
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/include/object_detector.h",
"new_path": "deploy/cpp/include/object_detector.h",
"diff": "#include \"include/utils.h\"\nusing namespace paddle_infer;\n-\nnamespace PaddleDetection {\n// Generate visualization colormap for each class\nstd::vector<int> GenerateColorMap(int num_class);\n// Visualiztion Detection Result\n-cv::Mat VisualizeResult(\n- const cv::Mat& img,\n+cv::Mat\n+VisualizeResult(const cv::Mat &img,\nconst std::vector<PaddleDetection::ObjectResult> &results,\nconst std::vector<std::string> &lables,\n- const std::vector<int>& colormap,\n- const bool is_rbox);\n+ const std::vector<int> &colormap, const bool is_rbox);\nclass ObjectDetector {\npublic:\nexplicit ObjectDetector(const std::string &model_dir,\nconst std::string &device = \"CPU\",\n- bool use_mkldnn = false,\n- int cpu_threads = 1,\n+ bool use_mkldnn = false, int cpu_threads = 1,\nconst std::string &run_mode = \"paddle\",\n- const int batch_size = 1,\n- const int gpu_id = 0,\n+ const int batch_size = 1, const int gpu_id = 0,\nconst int trt_min_shape = 1,\nconst int trt_max_shape = 1280,\nconst int trt_opt_shape = 640,\n@@ -78,15 +74,12 @@ class ObjectDetector {\n}\n// Load Paddle inference model\n- void LoadModel(const std::string& model_dir,\n- const int batch_size = 1,\n+ void LoadModel(const std::string &model_dir, const int batch_size = 1,\nconst std::string &run_mode = \"paddle\");\n// Run predictor\n- void Predict(const std::vector<cv::Mat> imgs,\n- const double threshold = 0.5,\n- const int warmup = 0,\n- const int repeats = 1,\n+ void Predict(const std::vector<cv::Mat> imgs, const double threshold = 0.5,\n+ const int warmup = 0, const int repeats = 1,\nstd::vector<PaddleDetection::ObjectResult> *result = nullptr,\nstd::vector<int> *bbox_num = nullptr,\nstd::vector<double> *times = nullptr);\n@@ -112,10 +105,14 @@ class ObjectDetector {\n// Postprocess result\nvoid Postprocess(const std::vector<cv::Mat> mats,\nstd::vector<PaddleDetection::ObjectResult> *result,\n- std::vector<int> bbox_num,\n- std::vector<float> output_data_,\n- std::vector<int> output_mask_data_,\n- bool is_rbox);\n+ std::vector<int> bbox_num, std::vector<float> output_data_,\n+ std::vector<int> output_mask_data_, bool is_rbox);\n+\n+ void SOLOv2Postprocess(\n+ const std::vector<cv::Mat> mats, std::vector<ObjectResult> *result,\n+ std::vector<int> *bbox_num, std::vector<int> out_bbox_num_data_,\n+ std::vector<int64_t> out_label_data_, std::vector<float> out_score_data_,\n+ std::vector<uint8_t> out_global_mask_data_, float threshold = 0.5);\nstd::shared_ptr<Predictor> predictor_;\nPreprocessor preprocessor_;\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/src/object_detector.cc",
"new_path": "deploy/cpp/src/object_detector.cc",
"diff": "@@ -41,17 +41,12 @@ void ObjectDetector::LoadModel(const std::string &model_dir,\n} else if (run_mode == \"trt_int8\") {\nprecision = paddle_infer::Config::Precision::kInt8;\n} else {\n- printf(\n- \"run_mode should be 'paddle', 'trt_fp32', 'trt_fp16' or \"\n+ printf(\"run_mode should be 'paddle', 'trt_fp32', 'trt_fp16' or \"\n\"'trt_int8'\");\n}\n// set tensorrt\n- config.EnableTensorRtEngine(1 << 30,\n- batch_size,\n- this->min_subgraph_size_,\n- precision,\n- false,\n- this->trt_calib_mode_);\n+ config.EnableTensorRtEngine(1 << 30, batch_size, this->min_subgraph_size_,\n+ precision, false, this->trt_calib_mode_);\n// set use dynamic shape\nif (this->use_dynamic_shape_) {\n@@ -69,8 +64,8 @@ void ObjectDetector::LoadModel(const std::string &model_dir,\nconst std::map<std::string, std::vector<int>> map_opt_input_shape = {\n{\"image\", opt_input_shape}};\n- config.SetTRTDynamicShapeInfo(\n- map_min_input_shape, map_max_input_shape, map_opt_input_shape);\n+ config.SetTRTDynamicShapeInfo(map_min_input_shape, map_max_input_shape,\n+ map_opt_input_shape);\nstd::cout << \"TensorRT dynamic shape enabled\" << std::endl;\n}\n}\n@@ -95,12 +90,11 @@ void ObjectDetector::LoadModel(const std::string &model_dir,\n}\n// Visualiztion MaskDetector results\n-cv::Mat VisualizeResult(\n- const cv::Mat &img,\n+cv::Mat\n+VisualizeResult(const cv::Mat &img,\nconst std::vector<PaddleDetection::ObjectResult> &results,\nconst std::vector<std::string> &lables,\n- const std::vector<int> &colormap,\n- const bool is_rbox = false) {\n+ const std::vector<int> &colormap, const bool is_rbox = false) {\ncv::Mat vis_img = img.clone();\nint img_h = vis_img.rows;\nint img_w = vis_img.cols;\n@@ -149,16 +143,10 @@ cv::Mat VisualizeResult(\nstd::vector<cv::Mat> contours;\ncv::Mat hierarchy;\nmask.convertTo(mask, CV_8U);\n- cv::findContours(\n- mask, contours, hierarchy, cv::RETR_CCOMP, cv::CHAIN_APPROX_SIMPLE);\n- cv::drawContours(colored_img,\n- contours,\n- -1,\n- roi_color,\n- -1,\n- cv::LINE_8,\n- hierarchy,\n- 100);\n+ cv::findContours(mask, contours, hierarchy, cv::RETR_CCOMP,\n+ cv::CHAIN_APPROX_SIMPLE);\n+ cv::drawContours(colored_img, contours, -1, roi_color, -1, cv::LINE_8,\n+ hierarchy, 100);\ncv::Mat debug_roi = vis_img;\ncolored_img = 0.4 * colored_img + 0.6 * vis_img;\n@@ -170,19 +158,13 @@ cv::Mat VisualizeResult(\norigin.y = results[i].rect[1];\n// Configure text background\n- cv::Rect text_back = cv::Rect(results[i].rect[0],\n- results[i].rect[1] - text_size.height,\n- text_size.width,\n- text_size.height);\n+ cv::Rect text_back =\n+ cv::Rect(results[i].rect[0], results[i].rect[1] - text_size.height,\n+ text_size.width, text_size.height);\n// Draw text, and background\ncv::rectangle(vis_img, text_back, roi_color, -1);\n- cv::putText(vis_img,\n- text,\n- origin,\n- font_face,\n- font_scale,\n- cv::Scalar(255, 255, 255),\n- thickness);\n+ cv::putText(vis_img, text, origin, font_face, font_scale,\n+ cv::Scalar(255, 255, 255), thickness);\n}\nreturn vis_img;\n}\n@@ -197,10 +179,8 @@ void ObjectDetector::Preprocess(const cv::Mat &ori_im) {\nvoid ObjectDetector::Postprocess(\nconst std::vector<cv::Mat> mats,\nstd::vector<PaddleDetection::ObjectResult> *result,\n- std::vector<int> bbox_num,\n- std::vector<float> output_data_,\n- std::vector<int> output_mask_data_,\n- bool is_rbox = false) {\n+ std::vector<int> bbox_num, std::vector<float> output_data_,\n+ std::vector<int> output_mask_data_, bool is_rbox = false) {\nresult->clear();\nint start_idx = 0;\nint total_num = std::accumulate(bbox_num.begin(), bbox_num.end(), 0);\n@@ -267,9 +247,81 @@ void ObjectDetector::Postprocess(\n}\n}\n+// This function is to convert output result from SOLOv2 to class ObjectResult\n+void ObjectDetector::SOLOv2Postprocess(\n+ const std::vector<cv::Mat> mats, std::vector<ObjectResult> *result,\n+ std::vector<int> *bbox_num, std::vector<int> out_bbox_num_data_,\n+ std::vector<int64_t> out_label_data_, std::vector<float> out_score_data_,\n+ std::vector<uint8_t> out_global_mask_data_, float threshold) {\n+\n+ for (int im_id = 0; im_id < mats.size(); im_id++) {\n+ cv::Mat mat = mats[im_id];\n+\n+ int valid_bbox_count = 0;\n+ for (int bbox_id = 0; bbox_id < out_bbox_num_data_[im_id]; ++bbox_id) {\n+ if (out_score_data_[bbox_id] >= threshold) {\n+ ObjectResult result_item;\n+ result_item.class_id = out_label_data_[bbox_id];\n+ result_item.confidence = out_score_data_[bbox_id];\n+ std::vector<int> global_mask;\n+\n+ for (int k = 0; k < mat.rows * mat.cols; ++k) {\n+ global_mask.push_back(static_cast<int>(\n+ out_global_mask_data_[k + bbox_id * mat.rows * mat.cols]));\n+ }\n+\n+ // find minimize bounding box from mask\n+ cv::Mat mask(mat.rows, mat.cols, CV_32SC1);\n+ std::memcpy(mask.data, global_mask.data(),\n+ global_mask.size() * sizeof(int));\n+\n+ cv::Mat mask_fp;\n+ cv::Mat rowSum;\n+ cv::Mat colSum;\n+ std::vector<float> sum_of_row(mat.rows);\n+ std::vector<float> sum_of_col(mat.cols);\n+\n+ mask.convertTo(mask_fp, CV_32FC1);\n+ cv::reduce(mask_fp, colSum, 0, CV_REDUCE_SUM, CV_32FC1);\n+ cv::reduce(mask_fp, rowSum, 1, CV_REDUCE_SUM, CV_32FC1);\n+\n+ for (int row_id = 0; row_id < mat.rows; ++row_id) {\n+ sum_of_row[row_id] = rowSum.at<float>(row_id, 0);\n+ }\n+\n+ for (int col_id = 0; col_id < mat.cols; ++col_id) {\n+ sum_of_col[col_id] = colSum.at<float>(0, col_id);\n+ }\n+\n+ auto it = std::find_if(sum_of_row.begin(), sum_of_row.end(),\n+ [](int x) { return x > 0.5; });\n+ int y1 = std::distance(sum_of_row.begin(), it);\n+\n+ auto it2 = std::find_if(sum_of_col.begin(), sum_of_col.end(),\n+ [](int x) { return x > 0.5; });\n+ int x1 = std::distance(sum_of_col.begin(), it2);\n+\n+ auto rit = std::find_if(sum_of_row.rbegin(), sum_of_row.rend(),\n+ [](int x) { return x > 0.5; });\n+ int y2 = std::distance(rit, sum_of_row.rend());\n+\n+ auto rit2 = std::find_if(sum_of_col.rbegin(), sum_of_col.rend(),\n+ [](int x) { return x > 0.5; });\n+ int x2 = std::distance(rit2, sum_of_col.rend());\n+\n+ result_item.rect = {x1, y1, x2, y2};\n+ result_item.mask = global_mask;\n+\n+ result->push_back(result_item);\n+ valid_bbox_count++;\n+ }\n+ }\n+ bbox_num->push_back(valid_bbox_count);\n+ }\n+}\n+\nvoid ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\n- const double threshold,\n- const int warmup,\n+ const double threshold, const int warmup,\nconst int repeats,\nstd::vector<PaddleDetection::ObjectResult> *result,\nstd::vector<int> *bbox_num,\n@@ -285,6 +337,11 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\nstd::vector<int> out_bbox_num_data_;\nstd::vector<int> out_mask_data_;\n+ // these parameters are for SOLOv2 output\n+ std::vector<float> out_score_data_;\n+ std::vector<uint8_t> out_global_mask_data_;\n+ std::vector<int64_t> out_label_data_;\n+\n// in_net img for each batch\nstd::vector<cv::Mat> in_net_img_all(batch_size);\n@@ -298,8 +355,8 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\nscale_factor_all[bs_idx * 2] = inputs_.scale_factor_[0];\nscale_factor_all[bs_idx * 2 + 1] = inputs_.scale_factor_[1];\n- in_data_all.insert(\n- in_data_all.end(), inputs_.im_data_.begin(), inputs_.im_data_.end());\n+ in_data_all.insert(in_data_all.end(), inputs_.im_data_.begin(),\n+ inputs_.im_data_.end());\n// collect in_net img\nin_net_img_all[bs_idx] = inputs_.in_net_im_;\n@@ -320,8 +377,8 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\npad_data.resize(rc * rh * rw);\nfloat *base = pad_data.data();\nfor (int i = 0; i < rc; ++i) {\n- cv::extractChannel(\n- pad_img, cv::Mat(rh, rw, CV_32FC1, base + i * rh * rw), i);\n+ cv::extractChannel(pad_img,\n+ cv::Mat(rh, rw, CV_32FC1, base + i * rh * rw), i);\n}\nin_data_all.insert(in_data_all.end(), pad_data.begin(), pad_data.end());\n}\n@@ -354,6 +411,64 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\nbool is_rbox = false;\nint reg_max = 7;\nint num_class = 80;\n+\n+ auto inference_start = std::chrono::steady_clock::now();\n+ if (config_.arch_ == \"SOLOv2\") {\n+ // warmup\n+ for (int i = 0; i < warmup; i++) {\n+ predictor_->Run();\n+ // Get output tensor\n+ auto output_names = predictor_->GetOutputNames();\n+ for (int j = 0; j < output_names.size(); j++) {\n+ auto output_tensor = predictor_->GetOutputHandle(output_names[j]);\n+ std::vector<int> output_shape = output_tensor->shape();\n+ int out_num = std::accumulate(output_shape.begin(), output_shape.end(),\n+ 1, std::multiplies<int>());\n+ if (j == 0) {\n+ out_bbox_num_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_bbox_num_data_.data());\n+ } else if (j == 1) {\n+ out_label_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_label_data_.data());\n+ } else if (j == 2) {\n+ out_score_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_score_data_.data());\n+ } else if (config_.mask_ && (j == 3)) {\n+ out_global_mask_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_global_mask_data_.data());\n+ }\n+ }\n+ }\n+\n+ inference_start = std::chrono::steady_clock::now();\n+ for (int i = 0; i < repeats; i++) {\n+ predictor_->Run();\n+ // Get output tensor\n+ out_tensor_list.clear();\n+ output_shape_list.clear();\n+ auto output_names = predictor_->GetOutputNames();\n+ for (int j = 0; j < output_names.size(); j++) {\n+ auto output_tensor = predictor_->GetOutputHandle(output_names[j]);\n+ std::vector<int> output_shape = output_tensor->shape();\n+ int out_num = std::accumulate(output_shape.begin(), output_shape.end(),\n+ 1, std::multiplies<int>());\n+ output_shape_list.push_back(output_shape);\n+ if (j == 0) {\n+ out_bbox_num_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_bbox_num_data_.data());\n+ } else if (j == 1) {\n+ out_label_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_label_data_.data());\n+ } else if (j == 2) {\n+ out_score_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_score_data_.data());\n+ } else if (config_.mask_ && (j == 3)) {\n+ out_global_mask_data_.resize(out_num);\n+ output_tensor->CopyToCpu(out_global_mask_data_.data());\n+ }\n+ }\n+ }\n+ } else {\n// warmup\nfor (int i = 0; i < warmup; i++) {\npredictor_->Run();\n@@ -362,8 +477,8 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\nfor (int j = 0; j < output_names.size(); j++) {\nauto output_tensor = predictor_->GetOutputHandle(output_names[j]);\nstd::vector<int> output_shape = output_tensor->shape();\n- int out_num = std::accumulate(\n- output_shape.begin(), output_shape.end(), 1, std::multiplies<int>());\n+ int out_num = std::accumulate(output_shape.begin(), output_shape.end(),\n+ 1, std::multiplies<int>());\nif (config_.mask_ && (j == 2)) {\nout_mask_data_.resize(out_num);\noutput_tensor->CopyToCpu(out_mask_data_.data());\n@@ -379,7 +494,7 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\n}\n}\n- auto inference_start = std::chrono::steady_clock::now();\n+ inference_start = std::chrono::steady_clock::now();\nfor (int i = 0; i < repeats; i++) {\npredictor_->Run();\n// Get output tensor\n@@ -389,8 +504,8 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\nfor (int j = 0; j < output_names.size(); j++) {\nauto output_tensor = predictor_->GetOutputHandle(output_names[j]);\nstd::vector<int> output_shape = output_tensor->shape();\n- int out_num = std::accumulate(\n- output_shape.begin(), output_shape.end(), 1, std::multiplies<int>());\n+ int out_num = std::accumulate(output_shape.begin(), output_shape.end(),\n+ 1, std::multiplies<int>());\noutput_shape_list.push_back(output_shape);\nif (config_.mask_ && (j == 2)) {\nout_mask_data_.resize(out_num);\n@@ -406,6 +521,8 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\n}\n}\n}\n+ }\n+\nauto inference_end = std::chrono::steady_clock::now();\nauto postprocess_start = std::chrono::steady_clock::now();\n// Postprocessing result\n@@ -420,30 +537,23 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs,\nreg_max = output_shape_list[i][2] / 4 - 1;\n}\nfloat *buffer = new float[out_tensor_list[i].size()];\n- memcpy(buffer,\n- &out_tensor_list[i][0],\n+ memcpy(buffer, &out_tensor_list[i][0],\nout_tensor_list[i].size() * sizeof(float));\noutput_data_list_.push_back(buffer);\n}\nPaddleDetection::PicoDetPostProcess(\n- result,\n- output_data_list_,\n- config_.fpn_stride_,\n- inputs_.im_shape_,\n- inputs_.scale_factor_,\n- config_.nms_info_[\"score_threshold\"].as<float>(),\n- config_.nms_info_[\"nms_threshold\"].as<float>(),\n- num_class,\n- reg_max);\n+ result, output_data_list_, config_.fpn_stride_, inputs_.im_shape_,\n+ inputs_.scale_factor_, config_.nms_info_[\"score_threshold\"].as<float>(),\n+ config_.nms_info_[\"nms_threshold\"].as<float>(), num_class, reg_max);\nbbox_num->push_back(result->size());\n+ } else if (config_.arch_ == \"SOLOv2\") {\n+ SOLOv2Postprocess(imgs, result, bbox_num, out_bbox_num_data_,\n+ out_label_data_, out_score_data_, out_global_mask_data_,\n+ threshold);\n} else {\nis_rbox = output_shape_list[0][output_shape_list[0].size() - 1] % 10 == 0;\n- Postprocess(imgs,\n- result,\n- out_bbox_num_data_,\n- out_tensor_list[0],\n- out_mask_data_,\n- is_rbox);\n+ Postprocess(imgs, result, out_bbox_num_data_, out_tensor_list[0],\n+ out_mask_data_, is_rbox);\nfor (int k = 0; k < out_bbox_num_data_.size(); k++) {\nint tmp = out_bbox_num_data_[k];\nbbox_num->push_back(tmp);\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add cpp infer support for solov2 (#6050)
|
499,339 |
25.05.2022 12:26:32
| -28,800 |
2dc058ad75021e2ad07aaaf16d9183a404fbc674
|
[dev] update amp, add amp_level
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -32,7 +32,6 @@ import paddle\nimport paddle.nn as nn\nimport paddle.distributed as dist\nfrom paddle.distributed import fleet\n-from paddle import amp\nfrom paddle.static import InputSpec\nfrom ppdet.optimizer import ModelEMA\n@@ -380,13 +379,21 @@ class Trainer(object):\nself.cfg['EvalDataset'] = self.cfg.EvalDataset = create(\n\"EvalDataset\")()\n+ model = self.model\nsync_bn = (getattr(self.cfg, 'norm_type', None) == 'sync_bn' and\nself.cfg.use_gpu and self._nranks > 1)\nif sync_bn:\n- self.model = paddle.nn.SyncBatchNorm.convert_sync_batchnorm(\n- self.model)\n+ model = paddle.nn.SyncBatchNorm.convert_sync_batchnorm(model)\n- model = self.model\n+ # enabel auto mixed precision mode\n+ use_amp = self.cfg.get('amp', False)\n+ amp_level = self.cfg.get('amp_level', 'O1')\n+ if use_amp:\n+ scaler = paddle.amp.GradScaler(\n+ enable=self.cfg.use_gpu or self.cfg.use_npu,\n+ init_loss_scaling=self.cfg.get('init_loss_scaling', 1024))\n+ model = paddle.amp.decorate(models=model, level=amp_level)\n+ # get distributed model\nif self.cfg.get('fleet', False):\nmodel = fleet.distributed_model(model)\nself.optimizer = fleet.distributed_optimizer(self.optimizer)\n@@ -394,13 +401,7 @@ class Trainer(object):\nfind_unused_parameters = self.cfg[\n'find_unused_parameters'] if 'find_unused_parameters' in self.cfg else False\nmodel = paddle.DataParallel(\n- self.model, find_unused_parameters=find_unused_parameters)\n-\n- # enabel auto mixed precision mode\n- if self.cfg.get('amp', False):\n- scaler = amp.GradScaler(\n- enable=self.cfg.use_gpu or self.cfg.use_npu,\n- init_loss_scaling=1024)\n+ model, find_unused_parameters=find_unused_parameters)\nself.status.update({\n'epoch_id': self.start_epoch,\n@@ -436,12 +437,12 @@ class Trainer(object):\nself._compose_callback.on_step_begin(self.status)\ndata['epoch_id'] = epoch_id\n- if self.cfg.get('amp', False):\n- with amp.auto_cast(enable=self.cfg.use_gpu):\n+ if use_amp:\n+ with paddle.amp.auto_cast(\n+ enable=self.cfg.use_gpu, level=amp_level):\n# model forward\noutputs = model(data)\nloss = outputs['loss']\n-\n# model backward\nscaled_loss = scaler.scale(loss)\nscaled_loss.backward()\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] update amp, add amp_level (#6054)
|
499,333 |
25.05.2022 14:06:12
| -28,800 |
51b7d4cf5e97b9d876ccf34b0a3503cee7b527eb
|
fit py36
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/source/dataset.py",
"new_path": "ppdet/data/source/dataset.py",
"diff": "@@ -23,7 +23,7 @@ from paddle.io import Dataset\nfrom ppdet.core.workspace import register, serializable\nfrom ppdet.utils.download import get_dataset_path\nimport copy\n-import ppdet.data.source as source\n+from ppdet.data import source\n@serializable\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fit py36 (#6056)
|
499,299 |
25.05.2022 21:21:58
| -28,800 |
d2f86f6eac798988b646efeb64bc94e3da545033
|
fix image infer error in pphuman
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipe_utils.py",
"new_path": "deploy/pphuman/pipe_utils.py",
"diff": "@@ -297,6 +297,8 @@ def crop_image_with_det(batch_input, det_res, thresh=0.3):\ncrop_res = []\nfor b_id, input in enumerate(batch_input):\nboxes_num_i = boxes_num[b_id]\n+ if boxes_num_i == 0:\n+ continue\nboxes_i = boxes[start_idx:start_idx + boxes_num_i, :]\nscore_i = score[start_idx:start_idx + boxes_num_i]\nres = []\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix image infer error in pphuman (#6060)
|
499,339 |
26.05.2022 12:19:19
| -28,800 |
2768e1ae40444f24e2ab1ff50e0931a50e9e67fd
|
[PPYOLOE] fix assigner bug
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/atss_assigner.py",
"new_path": "ppdet/modeling/assigners/atss_assigner.py",
"diff": "@@ -51,7 +51,6 @@ class ATSSAssigner(nn.Layer):\ndef _gather_topk_pyramid(self, gt2anchor_distances, num_anchors_list,\npad_gt_mask):\n- pad_gt_mask = pad_gt_mask.tile([1, 1, self.topk]).astype(paddle.bool)\ngt2anchor_distances_list = paddle.split(\ngt2anchor_distances, num_anchors_list, axis=-1)\nnum_anchors_index = np.cumsum(num_anchors_list).tolist()\n@@ -61,15 +60,12 @@ class ATSSAssigner(nn.Layer):\nfor distances, anchors_index in zip(gt2anchor_distances_list,\nnum_anchors_index):\nnum_anchors = distances.shape[-1]\n- topk_metrics, topk_idxs = paddle.topk(\n+ _, topk_idxs = paddle.topk(\ndistances, self.topk, axis=-1, largest=False)\ntopk_idxs_list.append(topk_idxs + anchors_index)\n- topk_idxs = paddle.where(pad_gt_mask, topk_idxs,\n- paddle.zeros_like(topk_idxs))\n- is_in_topk = F.one_hot(topk_idxs, num_anchors).sum(axis=-2)\n- is_in_topk = paddle.where(is_in_topk > 1,\n- paddle.zeros_like(is_in_topk), is_in_topk)\n- is_in_topk_list.append(is_in_topk.astype(gt2anchor_distances.dtype))\n+ is_in_topk = F.one_hot(topk_idxs, num_anchors).sum(\n+ axis=-2).astype(gt2anchor_distances.dtype)\n+ is_in_topk_list.append(is_in_topk * pad_gt_mask)\nis_in_topk_list = paddle.concat(is_in_topk_list, axis=-1)\ntopk_idxs_list = paddle.concat(topk_idxs_list, axis=-1)\nreturn is_in_topk_list, topk_idxs_list\n@@ -155,9 +151,8 @@ class ATSSAssigner(nn.Layer):\niou_threshold = iou_threshold.reshape([batch_size, num_max_boxes, -1])\niou_threshold = iou_threshold.mean(axis=-1, keepdim=True) + \\\niou_threshold.std(axis=-1, keepdim=True)\n- is_in_topk = paddle.where(\n- iou_candidates > iou_threshold.tile([1, 1, num_anchors]),\n- is_in_topk, paddle.zeros_like(is_in_topk))\n+ is_in_topk = paddle.where(iou_candidates > iou_threshold, is_in_topk,\n+ paddle.zeros_like(is_in_topk))\n# 6. check the positive sample's center in gt, [B, n, L]\nis_in_gts = check_points_inside_bboxes(anchor_centers, gt_bboxes)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"new_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"diff": "@@ -112,9 +112,7 @@ class TaskAlignedAssigner(nn.Layer):\n# select topk largest alignment metrics pred bbox as candidates\n# for each gt, [B, n, L]\nis_in_topk = gather_topk_anchors(\n- alignment_metrics * is_in_gts,\n- self.topk,\n- topk_mask=pad_gt_mask.tile([1, 1, self.topk]).astype(paddle.bool))\n+ alignment_metrics * is_in_gts, self.topk, topk_mask=pad_gt_mask)\n# select positive sample, [B, n, L]\nmask_positive = is_in_topk * is_in_gts * pad_gt_mask\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/utils.py",
"new_path": "ppdet/modeling/assigners/utils.py",
"diff": "@@ -88,7 +88,7 @@ def gather_topk_anchors(metrics, topk, largest=True, topk_mask=None, eps=1e-9):\nlargest (bool) : largest is a flag, if set to true,\nalgorithm will sort by descending order, otherwise sort by\nascending order. Default: True\n- topk_mask (Tensor, bool|None): shape[B, n, topk], ignore bbox mask,\n+ topk_mask (Tensor, float32): shape[B, n, 1], ignore bbox mask,\nDefault: None\neps (float): Default: 1e-9\nReturns:\n@@ -98,13 +98,11 @@ def gather_topk_anchors(metrics, topk, largest=True, topk_mask=None, eps=1e-9):\ntopk_metrics, topk_idxs = paddle.topk(\nmetrics, topk, axis=-1, largest=largest)\nif topk_mask is None:\n- topk_mask = (topk_metrics.max(axis=-1, keepdim=True) > eps).tile(\n- [1, 1, topk])\n- topk_idxs = paddle.where(topk_mask, topk_idxs, paddle.zeros_like(topk_idxs))\n- is_in_topk = F.one_hot(topk_idxs, num_anchors).sum(axis=-2)\n- is_in_topk = paddle.where(is_in_topk > 1,\n- paddle.zeros_like(is_in_topk), is_in_topk)\n- return is_in_topk.astype(metrics.dtype)\n+ topk_mask = (\n+ topk_metrics.max(axis=-1, keepdim=True) > eps).astype(metrics.dtype)\n+ is_in_topk = F.one_hot(topk_idxs, num_anchors).sum(\n+ axis=-2).astype(metrics.dtype)\n+ return is_in_topk * topk_mask\ndef check_points_inside_bboxes(points,\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[PPYOLOE] fix assigner bug (#6066)
|
499,333 |
31.05.2022 10:03:15
| -28,800 |
271347a39e009f60675e16256b6639625b7bfbab
|
fix aa order in rcnn enhance
|
[
{
"change_type": "MODIFY",
"old_path": "configs/rcnn_enhance/_base_/faster_rcnn_enhance_reader.yml",
"new_path": "configs/rcnn_enhance/_base_/faster_rcnn_enhance_reader.yml",
"diff": "@@ -2,9 +2,9 @@ worker_num: 2\nTrainReader:\nsample_transforms:\n- Decode: {}\n+ - AutoAugment: {autoaug_type: v1}\n- RandomResize: {target_size: [[384,1000], [416,1000], [448,1000], [480,1000], [512,1000], [544,1000], [576,1000], [608,1000], [640,1000], [672,1000]], interp: 2, keep_ratio: True}\n- RandomFlip: {prob: 0.5}\n- - AutoAugment: {autoaug_type: v1}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_transforms:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix aa order in rcnn enhance (#6095)
|
499,301 |
01.06.2022 11:08:24
| -28,800 |
ae27dcd95fb2e4fb4b39a2fe6ddbfdfa1b55af36
|
fix train dataset get_anno
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -218,13 +218,14 @@ class Trainer(object):\n# when do validation in train, annotation file should be get from\n# EvalReader instead of self.dataset(which is TrainReader)\n- anno_file = self.dataset.get_anno()\n- dataset = self.dataset\nif self.mode == 'train' and validate:\neval_dataset = self.cfg['EvalDataset']\neval_dataset.check_or_download_dataset()\nanno_file = eval_dataset.get_anno()\ndataset = eval_dataset\n+ else:\n+ dataset = self.dataset\n+ anno_file = dataset.get_anno()\nIouType = self.cfg['IouType'] if 'IouType' in self.cfg else 'bbox'\nif self.cfg.metric == \"COCO\":\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix train dataset get_anno (#6076)
|
499,304 |
02.06.2022 11:52:31
| -28,800 |
40d58469f42a253d995581fdcbd39e847cfa852f
|
fix picodet output name in openvino demo
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_openvino/picodet_openvino.h",
"new_path": "deploy/third_engine/demo_openvino/picodet_openvino.h",
"diff": "// limitations under the License.\n// reference from https://github.com/RangiLyu/nanodet/tree/main/demo_openvino\n-\n#ifndef _PICODET_OPENVINO_H_\n#define _PICODET_OPENVINO_H_\n-#include <string>\n-#include <opencv2/core.hpp>\n#include <inference_engine.hpp>\n+#include <opencv2/core.hpp>\n+#include <string>\n#define image_size 416\n-\n-typedef struct HeadInfo\n-{\n+typedef struct HeadInfo {\nstd::string cls_layer;\nstd::string dis_layer;\nint stride;\n} HeadInfo;\n-typedef struct BoxInfo\n-{\n+typedef struct BoxInfo {\nfloat x1;\nfloat y1;\nfloat x2;\n@@ -41,8 +37,7 @@ typedef struct BoxInfo\nint label;\n} BoxInfo;\n-class PicoDet\n-{\n+class PicoDet {\npublic:\nPicoDet(const char *param);\n@@ -54,25 +49,27 @@ public:\nstd::vector<HeadInfo> heads_info_{\n// cls_pred|dis_pred|stride\n- {\"save_infer_model/scale_0.tmp_1\", \"save_infer_model/scale_4.tmp_1\", 8},\n- {\"save_infer_model/scale_1.tmp_1\", \"save_infer_model/scale_5.tmp_1\", 16},\n- {\"save_infer_model/scale_2.tmp_1\", \"save_infer_model/scale_6.tmp_1\", 32},\n- {\"save_infer_model/scale_3.tmp_1\", \"save_infer_model/scale_7.tmp_1\", 64},\n+ {\"transpose_0.tmp_0\", \"transpose_1.tmp_0\", 8},\n+ {\"transpose_2.tmp_0\", \"transpose_3.tmp_0\", 16},\n+ {\"transpose_4.tmp_0\", \"transpose_5.tmp_0\", 32},\n+ {\"transpose_6.tmp_0\", \"transpose_7.tmp_0\", 64},\n};\n- std::vector<BoxInfo> detect(cv::Mat image, float score_threshold, float nms_threshold);\n+ std::vector<BoxInfo> detect(cv::Mat image, float score_threshold,\n+ float nms_threshold);\nprivate:\nvoid preprocess(cv::Mat &image, InferenceEngine::Blob::Ptr &blob);\n- void decode_infer(const float*& cls_pred, const float*& dis_pred, int stride, float threshold, std::vector<std::vector<BoxInfo>>& results);\n- BoxInfo disPred2Bbox(const float*& dfl_det, int label, float score, int x, int y, int stride);\n+ void decode_infer(const float *&cls_pred, const float *&dis_pred, int stride,\n+ float threshold,\n+ std::vector<std::vector<BoxInfo>> &results);\n+ BoxInfo disPred2Bbox(const float *&dfl_det, int label, float score, int x,\n+ int y, int stride);\nstatic void nms(std::vector<BoxInfo> &result, float nms_threshold);\nstd::string input_name_;\nint input_size_ = image_size;\nint num_class_ = 80;\nint reg_max_ = 7;\n-\n};\n-\n#endif\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_openvino_kpts/picodet_openvino.h",
"new_path": "deploy/third_engine/demo_openvino_kpts/picodet_openvino.h",
"diff": "@@ -48,25 +48,22 @@ class PicoDet {\nstd::vector<HeadInfo> heads_info_{\n// cls_pred|dis_pred|stride\n- {\"save_infer_model/scale_0.tmp_1\", \"save_infer_model/scale_4.tmp_1\", 8},\n- {\"save_infer_model/scale_1.tmp_1\", \"save_infer_model/scale_5.tmp_1\", 16},\n- {\"save_infer_model/scale_2.tmp_1\", \"save_infer_model/scale_6.tmp_1\", 32},\n- {\"save_infer_model/scale_3.tmp_1\", \"save_infer_model/scale_7.tmp_1\", 64},\n+ {\"transpose_0.tmp_0\", \"transpose_1.tmp_0\", 8},\n+ {\"transpose_2.tmp_0\", \"transpose_3.tmp_0\", 16},\n+ {\"transpose_4.tmp_0\", \"transpose_5.tmp_0\", 32},\n+ {\"transpose_6.tmp_0\", \"transpose_7.tmp_0\", 64},\n};\n- std::vector<BoxInfo> detect(cv::Mat image,\n- float score_threshold,\n+ std::vector<BoxInfo> detect(cv::Mat image, float score_threshold,\nfloat nms_threshold);\nprivate:\nvoid preprocess(cv::Mat &image, InferenceEngine::Blob::Ptr &blob);\n- void decode_infer(const float*& cls_pred,\n- const float*& dis_pred,\n- int stride,\n+ void decode_infer(const float *&cls_pred, const float *&dis_pred, int stride,\nfloat threshold,\nstd::vector<std::vector<BoxInfo>> &results);\n- BoxInfo disPred2Bbox(\n- const float*& dfl_det, int label, float score, int x, int y, int stride);\n+ BoxInfo disPred2Bbox(const float *&dfl_det, int label, float score, int x,\n+ int y, int stride);\nstatic void nms(std::vector<BoxInfo> &result, float nms_threshold);\nstd::string input_name_;\nint input_size_ = image_size;\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix picodet output name in openvino demo (#6114)
|
499,339 |
02.06.2022 16:37:27
| -28,800 |
bf7b674cfea28f6c10d718f137d73e2fe0d325ce
|
[TIPC] add onnx infer
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/serving/python/preprocess_ops.py",
"new_path": "deploy/serving/python/preprocess_ops.py",
"diff": "@@ -3,10 +3,14 @@ import cv2\nimport copy\n-def decode_image(im, img_info):\n+def decode_image(im):\nim = np.array(im)\n- img_info['im_shape'] = np.array(im.shape[:2], dtype=np.float32)\n- img_info['scale_factor'] = np.array([1., 1.], dtype=np.float32)\n+ img_info = {\n+ \"im_shape\": np.array(\n+ im.shape[:2], dtype=np.float32),\n+ \"scale_factor\": np.array(\n+ [1., 1.], dtype=np.float32)\n+ }\nreturn im, img_info\n@@ -399,16 +403,10 @@ class Compose:\nop_type = new_op_info.pop('type')\nself.transforms.append(eval(op_type)(**new_op_info))\n- self.im_info = {\n- 'scale_factor': np.array(\n- [1., 1.], dtype=np.float32),\n- 'im_shape': None\n- }\n-\ndef __call__(self, img):\n- img, self.im_info = decode_image(img, self.im_info)\n+ img, im_info = decode_image(img)\nfor t in self.transforms:\n- img, self.im_info = t(img, self.im_info)\n- inputs = copy.deepcopy(self.im_info)\n+ img, im_info = t(img, im_info)\n+ inputs = copy.deepcopy(im_info)\ninputs['image'] = img\nreturn inputs\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/serving/python/web_service.py",
"new_path": "deploy/serving/python/web_service.py",
"diff": "@@ -132,7 +132,7 @@ class PredictConfig(object):\nself.arch = yml_conf['arch']\nself.preprocess_infos = yml_conf['Preprocess']\nself.min_subgraph_size = yml_conf['min_subgraph_size']\n- self.labels = yml_conf['label_list']\n+ self.label_list = yml_conf['label_list']\nself.use_dynamic_shape = yml_conf['use_dynamic_shape']\nself.draw_threshold = yml_conf.get(\"draw_threshold\", 0.5)\nself.mask = yml_conf.get(\"mask\", False)\n@@ -189,8 +189,8 @@ class DetectorOp(Op):\nresult = {}\nfor k, num in zip(input_dict.keys(), bboxes_num):\nbbox = bboxes[idx:idx + num]\n- result[k] = self.parse_det_result(bbox, draw_threshold,\n- GLOBAL_VAR['model_config'].labels)\n+ result[k] = self.parse_det_result(\n+ bbox, draw_threshold, GLOBAL_VAR['model_config'].label_list)\nreturn result, None, \"\"\ndef collate_inputs(self, inputs):\n@@ -206,7 +206,7 @@ class DetectorOp(Op):\ndef parse_det_result(self, bbox, draw_threshold, label_list):\nresult = []\nfor line in bbox:\n- if line[1] > draw_threshold:\n+ if line[0] > -1 and line[1] > draw_threshold:\nresult.append(f\"{label_list[int(line[0])]} {line[1]} \"\nf\"{line[2]} {line[3]} {line[4]} {line[5]}\")\nreturn result\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_onnxruntime/infer_demo.py",
"new_path": "deploy/third_engine/demo_onnxruntime/infer_demo.py",
"diff": "@@ -55,7 +55,8 @@ class PicoDet():\norigin_shape = srcimg.shape[:2]\nim_scale_y = newh / float(origin_shape[0])\nim_scale_x = neww / float(origin_shape[1])\n- img_shape = np.array([[float(origin_shape[0]), float(origin_shape[1])]\n+ img_shape = np.array([\n+ [float(self.input_shape[0]), float(self.input_shape[1])]\n]).astype('float32')\nscale_factor = np.array([[im_scale_y, im_scale_x]]).astype('float32')\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/onnx/infer.py",
"diff": "+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing, software\n+# distributed under the License is distributed on an \"AS IS\" BASIS,\n+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+# See the License for the specific language governing permissions and\n+# limitations under the License.\n+\n+import os\n+import yaml\n+import argparse\n+import numpy as np\n+import glob\n+from onnxruntime import InferenceSession\n+\n+from preprocess import Compose\n+\n+# Global dictionary\n+SUPPORT_MODELS = {\n+ 'YOLO',\n+ 'RCNN',\n+ 'SSD',\n+ 'Face',\n+ 'FCOS',\n+ 'SOLOv2',\n+ 'TTFNet',\n+ 'S2ANet',\n+ 'JDE',\n+ 'FairMOT',\n+ 'DeepSORT',\n+ 'GFL',\n+ 'PicoDet',\n+ 'CenterNet',\n+ 'TOOD',\n+ 'RetinaNet',\n+ 'StrongBaseline',\n+ 'STGCN',\n+ 'YOLOX',\n+}\n+\n+parser = argparse.ArgumentParser(description=__doc__)\n+parser.add_argument(\"-c\", \"--config\", type=str, help=\"infer_cfg.yml\")\n+parser.add_argument(\n+ '--onnx_file', type=str, default=\"model.onnx\", help=\"onnx model file path\")\n+parser.add_argument(\"--image_dir\", type=str)\n+parser.add_argument(\"--image_file\", type=str)\n+\n+\n+def get_test_images(infer_dir, infer_img):\n+ \"\"\"\n+ Get image path list in TEST mode\n+ \"\"\"\n+ assert infer_img is not None or infer_dir is not None, \\\n+ \"--image_file or --image_dir should be set\"\n+ assert infer_img is None or os.path.isfile(infer_img), \\\n+ \"{} is not a file\".format(infer_img)\n+ assert infer_dir is None or os.path.isdir(infer_dir), \\\n+ \"{} is not a directory\".format(infer_dir)\n+\n+ # infer_img has a higher priority\n+ if infer_img and os.path.isfile(infer_img):\n+ return [infer_img]\n+\n+ images = set()\n+ infer_dir = os.path.abspath(infer_dir)\n+ assert os.path.isdir(infer_dir), \\\n+ \"infer_dir {} is not a directory\".format(infer_dir)\n+ exts = ['jpg', 'jpeg', 'png', 'bmp']\n+ exts += [ext.upper() for ext in exts]\n+ for ext in exts:\n+ images.update(glob.glob('{}/*.{}'.format(infer_dir, ext)))\n+ images = list(images)\n+\n+ assert len(images) > 0, \"no image found in {}\".format(infer_dir)\n+ print(\"Found {} inference images in total.\".format(len(images)))\n+\n+ return images\n+\n+\n+class PredictConfig(object):\n+ \"\"\"set config of preprocess, postprocess and visualize\n+ Args:\n+ model_dir (str): root path of infer_cfg.yml\n+ \"\"\"\n+\n+ def __init__(self, infer_config):\n+ # parsing Yaml config for Preprocess\n+ with open(infer_config) as f:\n+ yml_conf = yaml.safe_load(f)\n+ self.check_model(yml_conf)\n+ self.arch = yml_conf['arch']\n+ self.preprocess_infos = yml_conf['Preprocess']\n+ self.min_subgraph_size = yml_conf['min_subgraph_size']\n+ self.label_list = yml_conf['label_list']\n+ self.use_dynamic_shape = yml_conf['use_dynamic_shape']\n+ self.draw_threshold = yml_conf.get(\"draw_threshold\", 0.5)\n+ self.mask = yml_conf.get(\"mask\", False)\n+ self.tracker = yml_conf.get(\"tracker\", None)\n+ self.nms = yml_conf.get(\"NMS\", None)\n+ self.fpn_stride = yml_conf.get(\"fpn_stride\", None)\n+ if self.arch == 'RCNN' and yml_conf.get('export_onnx', False):\n+ print(\n+ 'The RCNN export model is used for ONNX and it only supports batch_size = 1'\n+ )\n+ self.print_config()\n+\n+ def check_model(self, yml_conf):\n+ \"\"\"\n+ Raises:\n+ ValueError: loaded model not in supported model type\n+ \"\"\"\n+ for support_model in SUPPORT_MODELS:\n+ if support_model in yml_conf['arch']:\n+ return True\n+ raise ValueError(\"Unsupported arch: {}, expect {}\".format(yml_conf[\n+ 'arch'], SUPPORT_MODELS))\n+\n+ def print_config(self):\n+ print('----------- Model Configuration -----------')\n+ print('%s: %s' % ('Model Arch', self.arch))\n+ print('%s: ' % ('Transform Order'))\n+ for op_info in self.preprocess_infos:\n+ print('--%s: %s' % ('transform op', op_info['type']))\n+ print('--------------------------------------------')\n+\n+\n+def predict_image(infer_config, predictor, img_list):\n+ # load preprocess transforms\n+ transforms = Compose(infer_config.preprocess_infos)\n+ # predict image\n+ for img_path in img_list:\n+ inputs = transforms(img_path)\n+ inputs_name = [var.name for var in predictor.get_inputs()]\n+ inputs = {k: inputs[k][None, ] for k in inputs_name}\n+\n+ outputs = predictor.run(output_names=None, input_feed=inputs)\n+\n+ print(\"ONNXRuntime predict: \")\n+ bboxes = np.array(outputs[0])\n+ for bbox in bboxes:\n+ if bbox[0] > -1 and bbox[1] > infer_config.draw_threshold:\n+ print(f\"{infer_config.label_list[int(bbox[0])]} {bbox[1]} \"\n+ f\"{bbox[2]} {bbox[3]} {bbox[4]} {bbox[5]}\")\n+\n+\n+if __name__ == '__main__':\n+ FLAGS = parser.parse_args()\n+ # load image list\n+ img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)\n+ # load predictor\n+ predictor = InferenceSession(FLAGS.onnx_file)\n+ # load infer config\n+ infer_config = PredictConfig(FLAGS.config)\n+\n+ predict_image(infer_config, predictor, img_list)\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/onnx/preprocess.py",
"diff": "+import numpy as np\n+import cv2\n+import copy\n+\n+\n+def decode_image(img_path):\n+ with open(img_path, 'rb') as f:\n+ im_read = f.read()\n+ data = np.frombuffer(im_read, dtype='uint8')\n+ im = cv2.imdecode(data, 1) # BGR mode, but need RGB mode\n+ im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)\n+ img_info = {\n+ \"im_shape\": np.array(\n+ im.shape[:2], dtype=np.float32),\n+ \"scale_factor\": np.array(\n+ [1., 1.], dtype=np.float32)\n+ }\n+ return im, img_info\n+\n+\n+class Resize(object):\n+ \"\"\"resize image by target_size and max_size\n+ Args:\n+ target_size (int): the target size of image\n+ keep_ratio (bool): whether keep_ratio or not, default true\n+ interp (int): method of resize\n+ \"\"\"\n+\n+ def __init__(self, target_size, keep_ratio=True, interp=cv2.INTER_LINEAR):\n+ if isinstance(target_size, int):\n+ target_size = [target_size, target_size]\n+ self.target_size = target_size\n+ self.keep_ratio = keep_ratio\n+ self.interp = interp\n+\n+ def __call__(self, im, im_info):\n+ \"\"\"\n+ Args:\n+ im (np.ndarray): image (np.ndarray)\n+ im_info (dict): info of image\n+ Returns:\n+ im (np.ndarray): processed image (np.ndarray)\n+ im_info (dict): info of processed image\n+ \"\"\"\n+ assert len(self.target_size) == 2\n+ assert self.target_size[0] > 0 and self.target_size[1] > 0\n+ im_channel = im.shape[2]\n+ im_scale_y, im_scale_x = self.generate_scale(im)\n+ im = cv2.resize(\n+ im,\n+ None,\n+ None,\n+ fx=im_scale_x,\n+ fy=im_scale_y,\n+ interpolation=self.interp)\n+ im_info['im_shape'] = np.array(im.shape[:2]).astype('float32')\n+ im_info['scale_factor'] = np.array(\n+ [im_scale_y, im_scale_x]).astype('float32')\n+ return im, im_info\n+\n+ def generate_scale(self, im):\n+ \"\"\"\n+ Args:\n+ im (np.ndarray): image (np.ndarray)\n+ Returns:\n+ im_scale_x: the resize ratio of X\n+ im_scale_y: the resize ratio of Y\n+ \"\"\"\n+ origin_shape = im.shape[:2]\n+ im_c = im.shape[2]\n+ if self.keep_ratio:\n+ im_size_min = np.min(origin_shape)\n+ im_size_max = np.max(origin_shape)\n+ target_size_min = np.min(self.target_size)\n+ target_size_max = np.max(self.target_size)\n+ im_scale = float(target_size_min) / float(im_size_min)\n+ if np.round(im_scale * im_size_max) > target_size_max:\n+ im_scale = float(target_size_max) / float(im_size_max)\n+ im_scale_x = im_scale\n+ im_scale_y = im_scale\n+ else:\n+ resize_h, resize_w = self.target_size\n+ im_scale_y = resize_h / float(origin_shape[0])\n+ im_scale_x = resize_w / float(origin_shape[1])\n+ return im_scale_y, im_scale_x\n+\n+\n+class NormalizeImage(object):\n+ \"\"\"normalize image\n+ Args:\n+ mean (list): im - mean\n+ std (list): im / std\n+ is_scale (bool): whether need im / 255\n+ is_channel_first (bool): if True: image shape is CHW, else: HWC\n+ \"\"\"\n+\n+ def __init__(self, mean, std, is_scale=True):\n+ self.mean = mean\n+ self.std = std\n+ self.is_scale = is_scale\n+\n+ def __call__(self, im, im_info):\n+ \"\"\"\n+ Args:\n+ im (np.ndarray): image (np.ndarray)\n+ im_info (dict): info of image\n+ Returns:\n+ im (np.ndarray): processed image (np.ndarray)\n+ im_info (dict): info of processed image\n+ \"\"\"\n+ im = im.astype(np.float32, copy=False)\n+ mean = np.array(self.mean)[np.newaxis, np.newaxis, :]\n+ std = np.array(self.std)[np.newaxis, np.newaxis, :]\n+\n+ if self.is_scale:\n+ im = im / 255.0\n+ im -= mean\n+ im /= std\n+ return im, im_info\n+\n+\n+class Permute(object):\n+ \"\"\"permute image\n+ Args:\n+ to_bgr (bool): whether convert RGB to BGR\n+ channel_first (bool): whether convert HWC to CHW\n+ \"\"\"\n+\n+ def __init__(self, ):\n+ super(Permute, self).__init__()\n+\n+ def __call__(self, im, im_info):\n+ \"\"\"\n+ Args:\n+ im (np.ndarray): image (np.ndarray)\n+ im_info (dict): info of image\n+ Returns:\n+ im (np.ndarray): processed image (np.ndarray)\n+ im_info (dict): info of processed image\n+ \"\"\"\n+ im = im.transpose((2, 0, 1)).copy()\n+ return im, im_info\n+\n+\n+class PadStride(object):\n+ \"\"\" padding image for model with FPN, instead PadBatch(pad_to_stride) in original config\n+ Args:\n+ stride (bool): model with FPN need image shape % stride == 0\n+ \"\"\"\n+\n+ def __init__(self, stride=0):\n+ self.coarsest_stride = stride\n+\n+ def __call__(self, im, im_info):\n+ \"\"\"\n+ Args:\n+ im (np.ndarray): image (np.ndarray)\n+ im_info (dict): info of image\n+ Returns:\n+ im (np.ndarray): processed image (np.ndarray)\n+ im_info (dict): info of processed image\n+ \"\"\"\n+ coarsest_stride = self.coarsest_stride\n+ if coarsest_stride <= 0:\n+ return im, im_info\n+ im_c, im_h, im_w = im.shape\n+ pad_h = int(np.ceil(float(im_h) / coarsest_stride) * coarsest_stride)\n+ pad_w = int(np.ceil(float(im_w) / coarsest_stride) * coarsest_stride)\n+ padding_im = np.zeros((im_c, pad_h, pad_w), dtype=np.float32)\n+ padding_im[:, :im_h, :im_w] = im\n+ return padding_im, im_info\n+\n+\n+class LetterBoxResize(object):\n+ def __init__(self, target_size):\n+ \"\"\"\n+ Resize image to target size, convert normalized xywh to pixel xyxy\n+ format ([x_center, y_center, width, height] -> [x0, y0, x1, y1]).\n+ Args:\n+ target_size (int|list): image target size.\n+ \"\"\"\n+ super(LetterBoxResize, self).__init__()\n+ if isinstance(target_size, int):\n+ target_size = [target_size, target_size]\n+ self.target_size = target_size\n+\n+ def letterbox(self, img, height, width, color=(127.5, 127.5, 127.5)):\n+ # letterbox: resize a rectangular image to a padded rectangular\n+ shape = img.shape[:2] # [height, width]\n+ ratio_h = float(height) / shape[0]\n+ ratio_w = float(width) / shape[1]\n+ ratio = min(ratio_h, ratio_w)\n+ new_shape = (round(shape[1] * ratio),\n+ round(shape[0] * ratio)) # [width, height]\n+ padw = (width - new_shape[0]) / 2\n+ padh = (height - new_shape[1]) / 2\n+ top, bottom = round(padh - 0.1), round(padh + 0.1)\n+ left, right = round(padw - 0.1), round(padw + 0.1)\n+\n+ img = cv2.resize(\n+ img, new_shape, interpolation=cv2.INTER_AREA) # resized, no border\n+ img = cv2.copyMakeBorder(\n+ img, top, bottom, left, right, cv2.BORDER_CONSTANT,\n+ value=color) # padded rectangular\n+ return img, ratio, padw, padh\n+\n+ def __call__(self, im, im_info):\n+ \"\"\"\n+ Args:\n+ im (np.ndarray): image (np.ndarray)\n+ im_info (dict): info of image\n+ Returns:\n+ im (np.ndarray): processed image (np.ndarray)\n+ im_info (dict): info of processed image\n+ \"\"\"\n+ assert len(self.target_size) == 2\n+ assert self.target_size[0] > 0 and self.target_size[1] > 0\n+ height, width = self.target_size\n+ h, w = im.shape[:2]\n+ im, ratio, padw, padh = self.letterbox(im, height=height, width=width)\n+\n+ new_shape = [round(h * ratio), round(w * ratio)]\n+ im_info['im_shape'] = np.array(new_shape, dtype=np.float32)\n+ im_info['scale_factor'] = np.array([ratio, ratio], dtype=np.float32)\n+ return im, im_info\n+\n+\n+class Pad(object):\n+ def __init__(self, size, fill_value=[114.0, 114.0, 114.0]):\n+ \"\"\"\n+ Pad image to a specified size.\n+ Args:\n+ size (list[int]): image target size\n+ fill_value (list[float]): rgb value of pad area, default (114.0, 114.0, 114.0)\n+ \"\"\"\n+ super(Pad, self).__init__()\n+ if isinstance(size, int):\n+ size = [size, size]\n+ self.size = size\n+ self.fill_value = fill_value\n+\n+ def __call__(self, im, im_info):\n+ im_h, im_w = im.shape[:2]\n+ h, w = self.size\n+ if h == im_h and w == im_w:\n+ im = im.astype(np.float32)\n+ return im, im_info\n+\n+ canvas = np.ones((h, w, 3), dtype=np.float32)\n+ canvas *= np.array(self.fill_value, dtype=np.float32)\n+ canvas[0:im_h, 0:im_w, :] = im.astype(np.float32)\n+ im = canvas\n+ return im, im_info\n+\n+\n+def rotate_point(pt, angle_rad):\n+ \"\"\"Rotate a point by an angle.\n+\n+ Args:\n+ pt (list[float]): 2 dimensional point to be rotated\n+ angle_rad (float): rotation angle by radian\n+\n+ Returns:\n+ list[float]: Rotated point.\n+ \"\"\"\n+ assert len(pt) == 2\n+ sn, cs = np.sin(angle_rad), np.cos(angle_rad)\n+ new_x = pt[0] * cs - pt[1] * sn\n+ new_y = pt[0] * sn + pt[1] * cs\n+ rotated_pt = [new_x, new_y]\n+\n+ return rotated_pt\n+\n+\n+def _get_3rd_point(a, b):\n+ \"\"\"To calculate the affine matrix, three pairs of points are required. This\n+ function is used to get the 3rd point, given 2D points a & b.\n+\n+ The 3rd point is defined by rotating vector `a - b` by 90 degrees\n+ anticlockwise, using b as the rotation center.\n+\n+ Args:\n+ a (np.ndarray): point(x,y)\n+ b (np.ndarray): point(x,y)\n+\n+ Returns:\n+ np.ndarray: The 3rd point.\n+ \"\"\"\n+ assert len(a) == 2\n+ assert len(b) == 2\n+ direction = a - b\n+ third_pt = b + np.array([-direction[1], direction[0]], dtype=np.float32)\n+\n+ return third_pt\n+\n+\n+def get_affine_transform(center,\n+ input_size,\n+ rot,\n+ output_size,\n+ shift=(0., 0.),\n+ inv=False):\n+ \"\"\"Get the affine transform matrix, given the center/scale/rot/output_size.\n+\n+ Args:\n+ center (np.ndarray[2, ]): Center of the bounding box (x, y).\n+ scale (np.ndarray[2, ]): Scale of the bounding box\n+ wrt [width, height].\n+ rot (float): Rotation angle (degree).\n+ output_size (np.ndarray[2, ]): Size of the destination heatmaps.\n+ shift (0-100%): Shift translation ratio wrt the width/height.\n+ Default (0., 0.).\n+ inv (bool): Option to inverse the affine transform direction.\n+ (inv=False: src->dst or inv=True: dst->src)\n+\n+ Returns:\n+ np.ndarray: The transform matrix.\n+ \"\"\"\n+ assert len(center) == 2\n+ assert len(output_size) == 2\n+ assert len(shift) == 2\n+ if not isinstance(input_size, (np.ndarray, list)):\n+ input_size = np.array([input_size, input_size], dtype=np.float32)\n+ scale_tmp = input_size\n+\n+ shift = np.array(shift)\n+ src_w = scale_tmp[0]\n+ dst_w = output_size[0]\n+ dst_h = output_size[1]\n+\n+ rot_rad = np.pi * rot / 180\n+ src_dir = rotate_point([0., src_w * -0.5], rot_rad)\n+ dst_dir = np.array([0., dst_w * -0.5])\n+\n+ src = np.zeros((3, 2), dtype=np.float32)\n+ src[0, :] = center + scale_tmp * shift\n+ src[1, :] = center + src_dir + scale_tmp * shift\n+ src[2, :] = _get_3rd_point(src[0, :], src[1, :])\n+\n+ dst = np.zeros((3, 2), dtype=np.float32)\n+ dst[0, :] = [dst_w * 0.5, dst_h * 0.5]\n+ dst[1, :] = np.array([dst_w * 0.5, dst_h * 0.5]) + dst_dir\n+ dst[2, :] = _get_3rd_point(dst[0, :], dst[1, :])\n+\n+ if inv:\n+ trans = cv2.getAffineTransform(np.float32(dst), np.float32(src))\n+ else:\n+ trans = cv2.getAffineTransform(np.float32(src), np.float32(dst))\n+\n+ return trans\n+\n+\n+class WarpAffine(object):\n+ \"\"\"Warp affine the image\n+ \"\"\"\n+\n+ def __init__(self,\n+ keep_res=False,\n+ pad=31,\n+ input_h=512,\n+ input_w=512,\n+ scale=0.4,\n+ shift=0.1):\n+ self.keep_res = keep_res\n+ self.pad = pad\n+ self.input_h = input_h\n+ self.input_w = input_w\n+ self.scale = scale\n+ self.shift = shift\n+\n+ def __call__(self, im, im_info):\n+ \"\"\"\n+ Args:\n+ im (np.ndarray): image (np.ndarray)\n+ im_info (dict): info of image\n+ Returns:\n+ im (np.ndarray): processed image (np.ndarray)\n+ im_info (dict): info of processed image\n+ \"\"\"\n+ img = cv2.cvtColor(im, cv2.COLOR_RGB2BGR)\n+\n+ h, w = img.shape[:2]\n+\n+ if self.keep_res:\n+ input_h = (h | self.pad) + 1\n+ input_w = (w | self.pad) + 1\n+ s = np.array([input_w, input_h], dtype=np.float32)\n+ c = np.array([w // 2, h // 2], dtype=np.float32)\n+\n+ else:\n+ s = max(h, w) * 1.0\n+ input_h, input_w = self.input_h, self.input_w\n+ c = np.array([w / 2., h / 2.], dtype=np.float32)\n+\n+ trans_input = get_affine_transform(c, s, 0, [input_w, input_h])\n+ img = cv2.resize(img, (w, h))\n+ inp = cv2.warpAffine(\n+ img, trans_input, (input_w, input_h), flags=cv2.INTER_LINEAR)\n+ return inp, im_info\n+\n+\n+class Compose:\n+ def __init__(self, transforms):\n+ self.transforms = []\n+ for op_info in transforms:\n+ new_op_info = op_info.copy()\n+ op_type = new_op_info.pop('type')\n+ self.transforms.append(eval(op_type)(**new_op_info))\n+\n+ def __call__(self, img_path):\n+ img, im_info = decode_image(img_path)\n+ for t in self.transforms:\n+ img, im_info = t(img, im_info)\n+ inputs = copy.deepcopy(im_info)\n+ inputs['image'] = img\n+ return inputs\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"new_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"diff": "@@ -93,7 +93,7 @@ class TaskAlignedAssigner(nn.Layer):\nreturn assigned_labels, assigned_bboxes, assigned_scores\n# compute iou between gt and pred bbox, [B, n, L]\n- ious = iou_similarity(gt_bboxes, pred_bboxes)\n+ ious = batch_iou_similarity(gt_bboxes, pred_bboxes)\n# gather pred bboxes class score\npred_scores = pred_scores.transpose([0, 2, 1])\nbatch_ind = paddle.arange(\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] add onnx infer (#6119)
|
499,339 |
07.06.2022 14:44:29
| -28,800 |
5c72d5a12b295d9242faaf37421e1b1257a00bc0
|
[TIPC] add dist train infer
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/ppyoloe/ppyoloe_crn_s_300e_coco_train_fleet_infer_python.txt",
"diff": "+===========================train_params===========================\n+model_name:ppyoloe_crn_s_300e_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=300\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=2\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_300e_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml -o\n+pact_train:tools/train.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_300e_coco.pdparams\n+norm_export:tools/export_model.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml -o\n+pact_export:tools/export_model.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --slim_config _template_kl_quant -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:gpu|cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1|2\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+--trt_max_shape:1600\n+===========================infer_benchmark_params===========================\n+numpy_infer_input:3x640x640.npy\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/prepare.sh",
"new_path": "test_tipc/prepare.sh",
"diff": "@@ -22,15 +22,15 @@ if [ ${MODE} = \"whole_train_whole_infer\" ];then\neval \"${python} ./dataset/coco/download_coco.py\"\nelif [ ${MODE} = \"cpp_infer\" ];then\n# download coco lite data\n- wget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/tipc/coco_tipc.tar\n+ wget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/tipc/coco_tipc.tar --no-check-certificate\ncd ./dataset/coco/ && tar -xvf coco_tipc.tar && mv -n coco_tipc/* .\nrm -rf coco_tipc/ && cd ../../\n# download wider_face lite data\n- wget -nc -P ./dataset/wider_face/ https://paddledet.bj.bcebos.com/data/tipc/wider_tipc.tar\n+ wget -nc -P ./dataset/wider_face/ https://paddledet.bj.bcebos.com/data/tipc/wider_tipc.tar --no-check-certificate\ncd ./dataset/wider_face/ && tar -xvf wider_tipc.tar && mv -n wider_tipc/* .\nrm -rf wider_tipc/ && cd ../../\n# download spine lite data\n- wget -nc -P ./dataset/spine_coco/ https://paddledet.bj.bcebos.com/data/tipc/spine_tipc.tar\n+ wget -nc -P ./dataset/spine_coco/ https://paddledet.bj.bcebos.com/data/tipc/spine_tipc.tar --no-check-certificate\ncd ./dataset/spine_coco/ && tar -xvf spine_tipc.tar && mv -n spine_tipc/* .\nrm -rf spine_tipc/ && cd ../../\nif [[ ${model_name} =~ \"s2anet\" ]]; then\n@@ -38,7 +38,7 @@ elif [ ${MODE} = \"cpp_infer\" ];then\ncd ../../\nfi\n# download mot lite data\n- wget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/tipc/mot_tipc.tar\n+ wget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/tipc/mot_tipc.tar --no-check-certificate\ncd ./dataset/mot/ && tar -xvf mot_tipc.tar && mv -n mot_tipc/* .\nrm -rf mot_tipc/ && cd ../../\n@@ -50,7 +50,7 @@ elif [ ${MODE} = \"cpp_infer\" ];then\necho \"################### Opencv already exists, skip downloading. ###################\"\nelse\nmkdir -p $(pwd)/deps && cd $(pwd)/deps\n- wget -c https://paddledet.bj.bcebos.com/data/opencv-3.4.16_gcc8.2_ffmpeg.tar.gz\n+ wget -c https://paddledet.bj.bcebos.com/data/opencv-3.4.16_gcc8.2_ffmpeg.tar.gz --no-check-certificate\ntar -xvf opencv-3.4.16_gcc8.2_ffmpeg.tar.gz && cd ../\necho \"################### Finish downloading opencv. ###################\"\nfi\n@@ -60,13 +60,13 @@ elif [ ${MODE} = \"benchmark_train\" ];then\npip install -U pip Cython\npip install -r requirements.txt\n# prepare lite benchmark coco data\n- wget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/coco_benchmark.tar\n+ wget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/coco_benchmark.tar --no-check-certificate\ncd ./dataset/coco/ && tar -xvf coco_benchmark.tar\nmv -u coco_benchmark/* ./\nls ./\ncd ../../\n# prepare lite benchmark mot data\n- wget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/mot_benchmark.tar\n+ wget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/mot_benchmark.tar --no-check-certificate\ncd ./dataset/mot/ && tar -xvf mot_benchmark.tar\nmv -u mot_benchmark/* ./\nls ./\n@@ -87,15 +87,15 @@ elif [ ${MODE} = \"serving_infer\" ];then\npython -m pip install paddlepaddle-gpu==2.2.2.post101 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html\nelse\n# download coco lite data\n- wget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/tipc/coco_tipc.tar\n+ wget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/tipc/coco_tipc.tar --no-check-certificate\ncd ./dataset/coco/ && tar -xvf coco_tipc.tar && mv -n coco_tipc/* .\nrm -rf coco_tipc/ && cd ../../\n# download wider_face lite data\n- wget -nc -P ./dataset/wider_face/ https://paddledet.bj.bcebos.com/data/tipc/wider_tipc.tar\n+ wget -nc -P ./dataset/wider_face/ https://paddledet.bj.bcebos.com/data/tipc/wider_tipc.tar --no-check-certificate\ncd ./dataset/wider_face/ && tar -xvf wider_tipc.tar && mv -n wider_tipc/* .\nrm -rf wider_tipc/ && cd ../../\n# download spine_coco lite data\n- wget -nc -P ./dataset/spine_coco/ https://paddledet.bj.bcebos.com/data/tipc/spine_tipc.tar\n+ wget -nc -P ./dataset/spine_coco/ https://paddledet.bj.bcebos.com/data/tipc/spine_tipc.tar --no-check-certificate\ncd ./dataset/spine_coco/ && tar -xvf spine_tipc.tar && mv -n spine_tipc/* .\nrm -rf spine_tipc/ && cd ../../\nif [[ ${model_name} =~ \"s2anet\" ]]; then\n@@ -103,7 +103,7 @@ else\ncd ../../\nfi\n# download mot lite data\n- wget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/tipc/mot_tipc.tar\n+ wget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/tipc/mot_tipc.tar --no-check-certificate\ncd ./dataset/mot/ && tar -xvf mot_tipc.tar && mv -n mot_tipc/* .\nrm -rf mot_tipc/ && cd ../../\nfi\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_train_inference_python.sh",
"new_path": "test_tipc/test_train_inference_python.sh",
"diff": "@@ -278,10 +278,16 @@ else\nset_save_model=$(func_set_params \"${save_model_key}\" \"${save_log}\")\nif [ ${#gpu} -le 2 ];then # train with cpu or single gpu\ncmd=\"${python} ${run_train} LearningRate.base_lr=0.0001 log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_train_params1} ${set_autocast}\"\n- elif [ ${#ips} -le 26 ];then # train with multi-gpu\n+ elif [ ${#ips} -le 15 ];then # train with multi-gpu\ncmd=\"${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train} log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_train_params1} ${set_autocast}\"\nelse # train with multi-machine\n- cmd=\"${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${set_use_gpu} ${run_train} log_iter=1 ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_train_params1} ${set_autocast}\"\n+ IFS=\",\"\n+ ips_array=(${ips})\n+ nodes=${#ips_array[@]}\n+ save_log=\"${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}\"\n+ IFS=\"|\"\n+ set_save_model=$(func_set_params \"${save_model_key}\" \"${save_log}\")\n+ cmd=\"${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_train_params1} ${set_autocast}\"\nfi\n# run train\neval $cmd\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] add dist train infer (#6141)
|
499,339 |
07.06.2022 20:43:18
| -28,800 |
18c2099aa7fb80c4cad3f30d9fec02154c0e2e5e
|
[TIPC] fix cpp infer bug, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_inference_cpp.sh",
"new_path": "test_tipc/test_inference_cpp.sh",
"diff": "@@ -129,11 +129,21 @@ else\nfi\n# build program\n-# TODO: set PADDLE_DIR and TENSORRT_ROOT\n-if [ -z $PADDLE_DIR ]; then\n- wget -nc https://paddle-inference-lib.bj.bcebos.com/2.2.2/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda11.1_cudnn8.1.1_trt7.2.3.4/paddle_inference.tgz --no-check-certificate\n+# TODO: set PADDLE_INFER_DIR and TENSORRT_ROOT\n+if [ -z $PADDLE_INFER_DIR ]; then\n+ Paddle_Infer_Link=$2\n+ if [ \"\" = \"$Paddle_Infer_Link\" ];then\n+ wget -nc https://paddle-inference-lib.bj.bcebos.com/2.2.2/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda10.1_cudnn7.6.5_trt6.0.1.5/paddle_inference.tgz --no-check-certificate\ntar zxf paddle_inference.tgz\n- PADDLE_DIR=$(pwd)/paddle_inference\n+ PADDLE_INFER_DIR=$(pwd)/paddle_inference\n+ else\n+ wget -nc $Paddle_Infer_Link --no-check-certificate\n+ tar zxf paddle_inference.tgz\n+ PADDLE_INFER_DIR=$(pwd)/paddle_inference\n+ if [ ! -d \"paddle_inference\" ]; then\n+ PADDLE_INFER_DIR=$(pwd)/paddle_inference_install_dir\n+ fi\n+ fi\nfi\nif [ -z $TENSORRT_ROOT ]; then\nTENSORRT_ROOT=/usr/local/TensorRT6-cuda10.1-cudnn7\n@@ -148,10 +158,10 @@ mkdir -p build\ncd ./build\ncmake .. \\\n-DWITH_GPU=ON \\\n- -DWITH_MKL=OFF \\\n+ -DWITH_MKL=ON \\\n-DWITH_TENSORRT=OFF \\\n-DPADDLE_LIB_NAME=libpaddle_inference \\\n- -DPADDLE_DIR=${PADDLE_DIR} \\\n+ -DPADDLE_DIR=${PADDLE_INFER_DIR} \\\n-DCUDA_LIB=${CUDA_LIB} \\\n-DCUDNN_LIB=${CUDNN_LIB} \\\n-DTENSORRT_LIB_DIR=${TENSORRT_LIB_DIR} \\\n@@ -160,13 +170,13 @@ cmake .. \\\n-DWITH_KEYPOINT=ON \\\n-DWITH_MOT=ON\n-make -j4\n+make -j8\ncd ../../../\necho \"################### build finished! ###################\"\n# set cuda device\n-GPUID=$2\n+GPUID=$3\nif [ ${#GPUID} -le 0 ];then\nenv=\" \"\nelse\n@@ -178,7 +188,6 @@ Count=0\nIFS=\"|\"\ninfer_quant_flag=(${cpp_infer_is_quant_list})\nfor infer_mode in ${cpp_infer_mode_list[*]}; do\n-\n# run export\ncase ${infer_mode} in\nnorm) run_export=${norm_export} ;;\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix cpp infer bug, test=document_fix (#6149)
|
499,304 |
07.06.2022 21:57:29
| -28,800 |
9c680e5b42d89d158ad17bd92a5956141c41dd62
|
fix is_crowd and difficult in Mosaic
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/transform/operators.py",
"new_path": "ppdet/data/transform/operators.py",
"diff": "@@ -3184,7 +3184,7 @@ class Mosaic(BaseOperator):\nif np.random.uniform(0., 1.) > self.prob:\nreturn sample[0]\n- mosaic_gt_bbox, mosaic_gt_class, mosaic_is_crowd = [], [], []\n+ mosaic_gt_bbox, mosaic_gt_class, mosaic_is_crowd, mosaic_difficult = [], [], [], []\ninput_h, input_w = self.input_dim\nyc = int(random.uniform(0.5 * input_h, 1.5 * input_h))\nxc = int(random.uniform(0.5 * input_w, 1.5 * input_w))\n@@ -3217,21 +3217,35 @@ class Mosaic(BaseOperator):\n_gt_bbox[:, 2] = scale * gt_bbox[:, 2] + padw\n_gt_bbox[:, 3] = scale * gt_bbox[:, 3] + padh\n- is_crowd = sp['is_crowd'] if 'is_crowd' in sp else np.zeros(\n- (len(_gt_bbox), 1), dtype=np.int32)\nmosaic_gt_bbox.append(_gt_bbox)\nmosaic_gt_class.append(sp['gt_class'])\n- mosaic_is_crowd.append(is_crowd)\n+ if 'is_crowd' in sp:\n+ mosaic_is_crowd.append(sp['is_crowd'])\n+ if 'difficult' in sp:\n+ mosaic_difficult.append(sp['difficult'])\n# 2. clip bbox and get mosaic_labels([gt_bbox, gt_class, is_crowd])\nif len(mosaic_gt_bbox):\nmosaic_gt_bbox = np.concatenate(mosaic_gt_bbox, 0)\nmosaic_gt_class = np.concatenate(mosaic_gt_class, 0)\n+ if mosaic_is_crowd:\nmosaic_is_crowd = np.concatenate(mosaic_is_crowd, 0)\nmosaic_labels = np.concatenate([\n- mosaic_gt_bbox, mosaic_gt_class.astype(mosaic_gt_bbox.dtype),\n+ mosaic_gt_bbox,\n+ mosaic_gt_class.astype(mosaic_gt_bbox.dtype),\nmosaic_is_crowd.astype(mosaic_gt_bbox.dtype)\n], 1)\n+ elif mosaic_difficult:\n+ mosaic_difficult = np.concatenate(mosaic_difficult, 0)\n+ mosaic_labels = np.concatenate([\n+ mosaic_gt_bbox,\n+ mosaic_gt_class.astype(mosaic_gt_bbox.dtype),\n+ mosaic_difficult.astype(mosaic_gt_bbox.dtype)\n+ ], 1)\n+ else:\n+ mosaic_labels = np.concatenate([\n+ mosaic_gt_bbox, mosaic_gt_class.astype(mosaic_gt_bbox.dtype)\n+ ], 1)\nif self.remove_outside_box:\n# for MOT dataset\nflag1 = mosaic_gt_bbox[:, 0] < 2 * input_w\n@@ -3268,11 +3282,23 @@ class Mosaic(BaseOperator):\nrandom.random() < self.mixup_prob):\nsample_mixup = sample[4]\nmixup_img = sample_mixup['image']\n+ if 'is_crowd' in sample_mixup:\ncp_labels = np.concatenate([\nsample_mixup['gt_bbox'],\nsample_mixup['gt_class'].astype(mosaic_labels.dtype),\nsample_mixup['is_crowd'].astype(mosaic_labels.dtype)\n], 1)\n+ elif 'difficult' in sample_mixup:\n+ cp_labels = np.concatenate([\n+ sample_mixup['gt_bbox'],\n+ sample_mixup['gt_class'].astype(mosaic_labels.dtype),\n+ sample_mixup['difficult'].astype(mosaic_labels.dtype)\n+ ], 1)\n+ else:\n+ cp_labels = np.concatenate([\n+ sample_mixup['gt_bbox'],\n+ sample_mixup['gt_class'].astype(mosaic_labels.dtype)\n+ ], 1)\nmosaic_img, mosaic_labels = self.mixup_augment(\nmosaic_img, mosaic_labels, self.input_dim, cp_labels, mixup_img)\n@@ -3284,7 +3310,10 @@ class Mosaic(BaseOperator):\nsample0['im_shape'][1] = sample0['w']\nsample0['gt_bbox'] = mosaic_labels[:, :4].astype(np.float32)\nsample0['gt_class'] = mosaic_labels[:, 4:5].astype(np.float32)\n+ if 'is_crowd' in sample[0]:\nsample0['is_crowd'] = mosaic_labels[:, 5:6].astype(np.float32)\n+ if 'difficult' in sample[0]:\n+ sample0['difficult'] = mosaic_labels[:, 5:6].astype(np.float32)\nreturn sample0\ndef mixup_augment(self, origin_img, origin_labels, input_dim, cp_labels,\n@@ -3351,9 +3380,12 @@ class Mosaic(BaseOperator):\ncp_bboxes_transformed_np[:, 1::2] - y_offset, 0, target_h)\ncls_labels = cp_labels[:, 4:5].copy()\n- crd_labels = cp_labels[:, 5:6].copy()\nbox_labels = cp_bboxes_transformed_np\n+ if cp_labels.shape[-1] == 6:\n+ crd_labels = cp_labels[:, 5:6].copy()\nlabels = np.hstack((box_labels, cls_labels, crd_labels))\n+ else:\n+ labels = np.hstack((box_labels, cls_labels))\nif self.remove_outside_box:\nlabels = labels[labels[:, 0] < target_w]\nlabels = labels[labels[:, 2] > 0]\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix is_crowd and difficult in Mosaic (#6150)
|
499,298 |
08.06.2022 18:42:41
| -28,800 |
170fa7d20d4a77dd9193a5c8ff160328431e98a0
|
update yolox cfg and doc
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "configs/yolox/yoloxv2_tiny_300e_coco.yml",
"diff": "+_BASE_: [\n+ 'yolox_tiny_300e_coco.yml'\n+]\n+weights: output/yoloxv2_tiny_300e_coco/model_final\n+\n+CSPDarkNet:\n+ arch: \"P5\" # using the same backbone of YOLOv5 releases v6.0 and later version\n+ return_idx: [2, 3, 4]\n+ depthwise: False\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update yolox cfg and doc (#6151)
|
499,298 |
08.06.2022 20:36:08
| -28,800 |
ed331ba25c3d8157f20209806c472d07e59b126a
|
remove ppdet ops roi pool align, add vision roi pool align
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/roi_extractor.py",
"new_path": "ppdet/modeling/heads/roi_extractor.py",
"diff": "@@ -29,7 +29,7 @@ class RoIAlign(object):\nRoI Align module\nFor more details, please refer to the document of roi_align in\n- in ppdet/modeing/ops.py\n+ in https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/vision/ops.py\nArgs:\nresolution (int): The output size, default 14\n@@ -76,12 +76,12 @@ class RoIAlign(object):\ndef __call__(self, feats, roi, rois_num):\nroi = paddle.concat(roi) if len(roi) > 1 else roi[0]\nif len(feats) == 1:\n- rois_feat = ops.roi_align(\n- feats[self.start_level],\n- roi,\n- self.resolution,\n- self.spatial_scale[0],\n- rois_num=rois_num,\n+ rois_feat = paddle.vision.ops.roi_align(\n+ x=feats[self.start_level],\n+ boxes=roi,\n+ boxes_num=rois_num,\n+ output_size=self.resolution,\n+ spatial_scale=self.spatial_scale[0],\naligned=self.aligned)\nelse:\noffset = 2\n@@ -96,13 +96,13 @@ class RoIAlign(object):\nrois_num=rois_num)\nrois_feat_list = []\nfor lvl in range(self.start_level, self.end_level + 1):\n- roi_feat = ops.roi_align(\n- feats[lvl],\n- rois_dist[lvl],\n- self.resolution,\n- self.spatial_scale[lvl],\n+ roi_feat = paddle.vision.ops.roi_align(\n+ x=feats[lvl],\n+ boxes=rois_dist[lvl],\n+ boxes_num=rois_num_dist[lvl],\n+ output_size=self.resolution,\n+ spatial_scale=self.spatial_scale[lvl],\nsampling_ratio=self.sampling_ratio,\n- rois_num=rois_num_dist[lvl],\naligned=self.aligned)\nrois_feat_list.append(roi_feat)\nrois_feat_shuffle = paddle.concat(rois_feat_list)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/ops.py",
"new_path": "ppdet/modeling/ops.py",
"diff": "@@ -23,8 +23,6 @@ from paddle import in_dynamic_mode\nfrom paddle.common_ops_import import Variable, LayerHelper, check_variable_and_dtype, check_type, check_dtype\n__all__ = [\n- 'roi_pool',\n- 'roi_align',\n'prior_box',\n'generate_proposals',\n'box_coder',\n@@ -117,215 +115,6 @@ def batch_norm(ch,\nreturn norm_layer\n-@paddle.jit.not_to_static\n-def roi_pool(input,\n- rois,\n- output_size,\n- spatial_scale=1.0,\n- rois_num=None,\n- name=None):\n- \"\"\"\n-\n- This operator implements the roi_pooling layer.\n- Region of interest pooling (also known as RoI pooling) is to perform max pooling on inputs of nonuniform sizes to obtain fixed-size feature maps (e.g. 7*7).\n-\n- The operator has three steps:\n-\n- 1. Dividing each region proposal into equal-sized sections with output_size(h, w);\n- 2. Finding the largest value in each section;\n- 3. Copying these max values to the output buffer.\n-\n- For more information, please refer to https://stackoverflow.com/questions/43430056/what-is-roi-layer-in-fast-rcnn\n-\n- Args:\n- input (Tensor): Input feature, 4D-Tensor with the shape of [N,C,H,W],\n- where N is the batch size, C is the input channel, H is Height, W is weight.\n- The data type is float32 or float64.\n- rois (Tensor): ROIs (Regions of Interest) to pool over.\n- 2D-Tensor or 2D-LoDTensor with the shape of [num_rois,4], the lod level is 1.\n- Given as [[x1, y1, x2, y2], ...], (x1, y1) is the top left coordinates,\n- and (x2, y2) is the bottom right coordinates.\n- output_size (int or tuple[int, int]): The pooled output size(h, w), data type is int32. If int, h and w are both equal to output_size.\n- spatial_scale (float, optional): Multiplicative spatial scale factor to translate ROI coords from their input scale to the scale used when pooling. Default: 1.0\n- rois_num (Tensor): The number of RoIs in each image. Default: None\n- name(str, optional): For detailed information, please refer\n- to :ref:`api_guide_Name`. Usually name is no need to set and\n- None by default.\n-\n-\n- Returns:\n- Tensor: The pooled feature, 4D-Tensor with the shape of [num_rois, C, output_size[0], output_size[1]].\n-\n-\n- Examples:\n-\n- .. code-block:: python\n-\n- import paddle\n- from ppdet.modeling import ops\n- paddle.enable_static()\n-\n- x = paddle.static.data(\n- name='data', shape=[None, 256, 32, 32], dtype='float32')\n- rois = paddle.static.data(\n- name='rois', shape=[None, 4], dtype='float32')\n- rois_num = paddle.static.data(name='rois_num', shape=[None], dtype='int32')\n-\n- pool_out = ops.roi_pool(\n- input=x,\n- rois=rois,\n- output_size=(1, 1),\n- spatial_scale=1.0,\n- rois_num=rois_num)\n- \"\"\"\n- check_type(output_size, 'output_size', (int, tuple), 'roi_pool')\n- if isinstance(output_size, int):\n- output_size = (output_size, output_size)\n-\n- pooled_height, pooled_width = output_size\n- if in_dynamic_mode():\n- assert rois_num is not None, \"rois_num should not be None in dygraph mode.\"\n- pool_out, argmaxes = _C_ops.roi_pool(\n- input, rois, rois_num, \"pooled_height\", pooled_height,\n- \"pooled_width\", pooled_width, \"spatial_scale\", spatial_scale)\n- return pool_out, argmaxes\n-\n- else:\n- check_variable_and_dtype(input, 'input', ['float32'], 'roi_pool')\n- check_variable_and_dtype(rois, 'rois', ['float32'], 'roi_pool')\n- helper = LayerHelper('roi_pool', **locals())\n- dtype = helper.input_dtype()\n- pool_out = helper.create_variable_for_type_inference(dtype)\n- argmaxes = helper.create_variable_for_type_inference(dtype='int32')\n-\n- inputs = {\n- \"X\": input,\n- \"ROIs\": rois,\n- }\n- if rois_num is not None:\n- inputs['RoisNum'] = rois_num\n- helper.append_op(\n- type=\"roi_pool\",\n- inputs=inputs,\n- outputs={\"Out\": pool_out,\n- \"Argmax\": argmaxes},\n- attrs={\n- \"pooled_height\": pooled_height,\n- \"pooled_width\": pooled_width,\n- \"spatial_scale\": spatial_scale\n- })\n- return pool_out, argmaxes\n-\n-\n-@paddle.jit.not_to_static\n-def roi_align(input,\n- rois,\n- output_size,\n- spatial_scale=1.0,\n- sampling_ratio=-1,\n- rois_num=None,\n- aligned=True,\n- name=None):\n- \"\"\"\n-\n- Region of interest align (also known as RoI align) is to perform\n- bilinear interpolation on inputs of nonuniform sizes to obtain\n- fixed-size feature maps (e.g. 7*7)\n-\n- Dividing each region proposal into equal-sized sections with\n- the pooled_width and pooled_height. Location remains the origin\n- result.\n-\n- In each ROI bin, the value of the four regularly sampled locations\n- are computed directly through bilinear interpolation. The output is\n- the mean of four locations.\n- Thus avoid the misaligned problem.\n-\n- Args:\n- input (Tensor): Input feature, 4D-Tensor with the shape of [N,C,H,W],\n- where N is the batch size, C is the input channel, H is Height, W is weight.\n- The data type is float32 or float64.\n- rois (Tensor): ROIs (Regions of Interest) to pool over.It should be\n- a 2-D Tensor or 2-D LoDTensor of shape (num_rois, 4), the lod level is 1.\n- The data type is float32 or float64. Given as [[x1, y1, x2, y2], ...],\n- (x1, y1) is the top left coordinates, and (x2, y2) is the bottom right coordinates.\n- output_size (int or tuple[int, int]): The pooled output size(h, w), data type is int32. If int, h and w are both equal to output_size.\n- spatial_scale (float32, optional): Multiplicative spatial scale factor to translate ROI coords\n- from their input scale to the scale used when pooling. Default: 1.0\n- sampling_ratio(int32, optional): number of sampling points in the interpolation grid.\n- If <=0, then grid points are adaptive to roi_width and pooled_w, likewise for height. Default: -1\n- rois_num (Tensor): The number of RoIs in each image. Default: None\n- name(str, optional): For detailed information, please refer\n- to :ref:`api_guide_Name`. Usually name is no need to set and\n- None by default.\n-\n- Returns:\n- Tensor:\n-\n- Output: The output of ROIAlignOp is a 4-D tensor with shape (num_rois, channels, pooled_h, pooled_w). The data type is float32 or float64.\n-\n-\n- Examples:\n- .. code-block:: python\n-\n- import paddle\n- from ppdet.modeling import ops\n- paddle.enable_static()\n-\n- x = paddle.static.data(\n- name='data', shape=[None, 256, 32, 32], dtype='float32')\n- rois = paddle.static.data(\n- name='rois', shape=[None, 4], dtype='float32')\n- rois_num = paddle.static.data(name='rois_num', shape=[None], dtype='int32')\n- align_out = ops.roi_align(input=x,\n- rois=rois,\n- output_size=(7, 7),\n- spatial_scale=0.5,\n- sampling_ratio=-1,\n- rois_num=rois_num)\n- \"\"\"\n- check_type(output_size, 'output_size', (int, tuple), 'roi_align')\n- if isinstance(output_size, int):\n- output_size = (output_size, output_size)\n-\n- pooled_height, pooled_width = output_size\n-\n- if in_dynamic_mode():\n- assert rois_num is not None, \"rois_num should not be None in dygraph mode.\"\n- align_out = _C_ops.roi_align(\n- input, rois, rois_num, \"pooled_height\", pooled_height,\n- \"pooled_width\", pooled_width, \"spatial_scale\", spatial_scale,\n- \"sampling_ratio\", sampling_ratio, \"aligned\", aligned)\n- return align_out\n-\n- else:\n- check_variable_and_dtype(input, 'input', ['float32', 'float64'],\n- 'roi_align')\n- check_variable_and_dtype(rois, 'rois', ['float32', 'float64'],\n- 'roi_align')\n- helper = LayerHelper('roi_align', **locals())\n- dtype = helper.input_dtype()\n- align_out = helper.create_variable_for_type_inference(dtype)\n- inputs = {\n- \"X\": input,\n- \"ROIs\": rois,\n- }\n- if rois_num is not None:\n- inputs['RoisNum'] = rois_num\n- helper.append_op(\n- type=\"roi_align\",\n- inputs=inputs,\n- outputs={\"Out\": align_out},\n- attrs={\n- \"pooled_height\": pooled_height,\n- \"pooled_width\": pooled_width,\n- \"spatial_scale\": spatial_scale,\n- \"sampling_ratio\": sampling_ratio,\n- \"aligned\": aligned,\n- })\n- return align_out\n-\n-\n@paddle.jit.not_to_static\ndef distribute_fpn_proposals(fpn_rois,\nmin_level,\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/tests/test_ops.py",
"new_path": "ppdet/modeling/tests/test_ops.py",
"diff": "@@ -128,11 +128,11 @@ class TestROIAlign(LayerTest):\nrois_num = paddle.static.data(\nname='rois_num', shape=[None], dtype='int32')\n- output = ops.roi_align(\n- input=inputs,\n- rois=rois,\n- output_size=output_size,\n- rois_num=rois_num)\n+ output = paddle.vision.ops.roi_align(\n+ x=inputs,\n+ boxes=rois,\n+ boxes_num=rois_num,\n+ output_size=output_size)\noutput_np, = self.get_static_graph_result(\nfeed={\n'inputs': inputs_np,\n@@ -147,11 +147,11 @@ class TestROIAlign(LayerTest):\nrois_dy = paddle.to_tensor(rois_np)\nrois_num_dy = paddle.to_tensor(rois_num_np)\n- output_dy = ops.roi_align(\n- input=inputs_dy,\n- rois=rois_dy,\n- output_size=output_size,\n- rois_num=rois_num_dy)\n+ output_dy = paddle.vision.ops.roi_align(\n+ x=inputs_dy,\n+ boxes=rois_dy,\n+ boxes_num=rois_num_dy,\n+ output_size=output_size)\noutput_dy_np = output_dy.numpy()\nself.assertTrue(np.array_equal(output_np, output_dy_np))\n@@ -164,7 +164,7 @@ class TestROIAlign(LayerTest):\nname='data_error', shape=[10, 4], dtype='int32', lod_level=1)\nself.assertRaises(\nTypeError,\n- ops.roi_align,\n+ paddle.vision.ops.roi_align,\ninput=inputs,\nrois=rois,\noutput_size=(7, 7))\n@@ -188,11 +188,11 @@ class TestROIPool(LayerTest):\nrois_num = paddle.static.data(\nname='rois_num', shape=[None], dtype='int32')\n- output, _ = ops.roi_pool(\n- input=inputs,\n- rois=rois,\n- output_size=output_size,\n- rois_num=rois_num)\n+ output = paddle.vision.ops.roi_pool(\n+ x=inputs,\n+ boxes=rois,\n+ boxes_num=rois_num,\n+ output_size=output_size)\noutput_np, = self.get_static_graph_result(\nfeed={\n'inputs': inputs_np,\n@@ -207,11 +207,11 @@ class TestROIPool(LayerTest):\nrois_dy = paddle.to_tensor(rois_np)\nrois_num_dy = paddle.to_tensor(rois_num_np)\n- output_dy, _ = ops.roi_pool(\n- input=inputs_dy,\n- rois=rois_dy,\n- output_size=output_size,\n- rois_num=rois_num_dy)\n+ output_dy = paddle.vision.ops.roi_pool(\n+ x=inputs_dy,\n+ boxes=rois_dy,\n+ boxes_num=rois_num_dy,\n+ output_size=output_size)\noutput_dy_np = output_dy.numpy()\nself.assertTrue(np.array_equal(output_np, output_dy_np))\n@@ -224,7 +224,7 @@ class TestROIPool(LayerTest):\nname='data_error', shape=[10, 4], dtype='int32', lod_level=1)\nself.assertRaises(\nTypeError,\n- ops.roi_pool,\n+ paddle.vision.ops.roi_pool,\ninput=inputs,\nrois=rois,\noutput_size=(7, 7))\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
remove ppdet ops roi pool align, add vision roi pool align (#6154)
|
499,333 |
09.06.2022 17:28:16
| -28,800 |
f18e57984b2953320c5317eabcffca03080b36ed
|
keep device in export
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/export_utils.py",
"new_path": "ppdet/engine/export_utils.py",
"diff": "@@ -58,7 +58,9 @@ MOT_ARCH = ['DeepSORT', 'JDE', 'FairMOT', 'ByteTrack']\ndef _prune_input_spec(input_spec, program, targets):\n# try to prune static program to figure out pruned input spec\n# so we perform following operations in static mode\n+ device = paddle.get_device()\npaddle.enable_static()\n+ paddle.set_device(device)\npruned_input_spec = [{}]\nprogram = program.clone()\nprogram = program._prune(targets=targets)\n@@ -69,7 +71,7 @@ def _prune_input_spec(input_spec, program, targets):\npruned_input_spec[0][name] = spec\nexcept Exception:\npass\n- paddle.disable_static()\n+ paddle.disable_static(place=device)\nreturn pruned_input_spec\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
keep device in export (#6157)
|
499,333 |
09.06.2022 18:06:56
| -28,800 |
636b8c4794202470e310c8af07c9a632b7e90671
|
fix mask rcnn in eval when num_classes is 1
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/mask_head.py",
"new_path": "ppdet/modeling/heads/mask_head.py",
"diff": "@@ -222,7 +222,7 @@ class MaskHead(nn.Layer):\nmask_feat = self.head(rois_feat)\nmask_logit = self.mask_fcn_logits(mask_feat)\nif self.num_classes == 1:\n- mask_out = F.sigmoid(mask_logit)\n+ mask_out = F.sigmoid(mask_logit)[:, 0, :, :]\nelse:\nnum_masks = paddle.shape(mask_logit)[0]\nindex = paddle.arange(num_masks).cast('int32')\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix mask rcnn in eval when num_classes is 1 (#6168)
|
499,339 |
10.06.2022 14:04:42
| -28,800 |
ff62e6ff4abbbdcff02834c5d058dce6addb34d2
|
[TIPC] fix fleet train shell name, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/keypoint/tinypose_128x96_model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/keypoint/tinypose_128x96_model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt",
"diff": "@@ -20,7 +20,7 @@ inference:./deploy/cpp/build/main\n--device:gpu|cpu\n--use_mkldnn:False\n--cpu_threads:4\n---batch_size:1|2\n+--batch_size:1\n--use_tensorrt:null\n--run_mode:paddle\n--model_dir_keypoint:\n"
},
{
"change_type": "RENAME",
"old_path": "test_tipc/configs/ppyoloe/ppyoloe_crn_s_300e_coco_train_fleet_infer_python.txt",
"new_path": "test_tipc/configs/ppyoloe/ppyoloe_crn_s_300e_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": ""
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix fleet train shell name, test=document_fix (#6176)
|
499,363 |
14.06.2022 14:07:37
| -28,800 |
1c4da10b6c836f7f0b74c0847afcbca6d0c3ef30
|
Scale frames before fight action recognition
* Scale frames before fight action recognition
* put
short_size = self.cfg["VIDEO_ACTION"]["short_size"]
scale = Scale(short_size)
out of while
* change class name from Scale to ShortSizeScale
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -42,7 +42,7 @@ from python.action_utils import KeyPointBuff, SkeletonActionVisualHelper\nfrom pipe_utils import argsparser, print_arguments, merge_cfg, PipeTimer\nfrom pipe_utils import get_test_images, crop_image_with_det, crop_image_with_mot, parse_mot_res, parse_mot_keypoint\n-from python.preprocess import decode_image\n+from python.preprocess import decode_image, ShortSizeScale\nfrom python.visualize import visualize_box_mask, visualize_attr, visualize_pose, visualize_action\nfrom pptracking.python.mot_sde_infer import SDE_Detector\n@@ -554,6 +554,10 @@ class PipePredictor(object):\nvideo_action_imgs = []\n+ if self.with_video_action:\n+ short_size = self.cfg[\"VIDEO_ACTION\"][\"short_size\"]\n+ scale = ShortSizeScale(short_size)\n+\nwhile (1):\nif frame_id % 10 == 0:\nprint('frame id: ', frame_id)\n@@ -705,7 +709,9 @@ class PipePredictor(object):\n# collect frames\nif frame_id % sample_freq == 0:\n- video_action_imgs.append(frame)\n+ # Scale image\n+ scaled_img = scale(frame)\n+ video_action_imgs.append(scaled_img)\n# the number of collected frames is enough to predict video action\nif len(video_action_imgs) == frame_len:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/preprocess.py",
"new_path": "deploy/python/preprocess.py",
"diff": "import cv2\nimport numpy as np\nfrom keypoint_preprocess import get_affine_transform\n+from PIL import Image\ndef decode_image(im_file, im_info):\n@@ -106,6 +107,95 @@ class Resize(object):\nreturn im_scale_y, im_scale_x\n+class ShortSizeScale(object):\n+ \"\"\"\n+ Scale images by short size.\n+ Args:\n+ short_size(float | int): Short size of an image will be scaled to the short_size.\n+ fixed_ratio(bool): Set whether to zoom according to a fixed ratio. default: True\n+ do_round(bool): Whether to round up when calculating the zoom ratio. default: False\n+ backend(str): Choose pillow or cv2 as the graphics processing backend. default: 'pillow'\n+ \"\"\"\n+\n+ def __init__(self,\n+ short_size,\n+ fixed_ratio=True,\n+ keep_ratio=None,\n+ do_round=False,\n+ backend='pillow'):\n+ self.short_size = short_size\n+ assert (fixed_ratio and not keep_ratio) or (\n+ not fixed_ratio\n+ ), \"fixed_ratio and keep_ratio cannot be true at the same time\"\n+ self.fixed_ratio = fixed_ratio\n+ self.keep_ratio = keep_ratio\n+ self.do_round = do_round\n+\n+ assert backend in [\n+ 'pillow', 'cv2'\n+ ], \"Scale's backend must be pillow or cv2, but get {backend}\"\n+\n+ self.backend = backend\n+\n+ def __call__(self, img):\n+ \"\"\"\n+ Performs resize operations.\n+ Args:\n+ img (PIL.Image): a PIL.Image.\n+ return:\n+ resized_img: a PIL.Image after scaling.\n+ \"\"\"\n+\n+ result_img = None\n+\n+ if isinstance(img, np.ndarray):\n+ h, w, _ = img.shape\n+ elif isinstance(img, Image.Image):\n+ w, h = img.size\n+ else:\n+ raise NotImplementedError\n+\n+ if w <= h:\n+ ow = self.short_size\n+ if self.fixed_ratio: # default is True\n+ oh = int(self.short_size * 4.0 / 3.0)\n+ elif not self.keep_ratio: # no\n+ oh = self.short_size\n+ else:\n+ scale_factor = self.short_size / w\n+ oh = int(h * float(scale_factor) +\n+ 0.5) if self.do_round else int(h * self.short_size / w)\n+ ow = int(w * float(scale_factor) +\n+ 0.5) if self.do_round else int(w * self.short_size / h)\n+ else:\n+ oh = self.short_size\n+ if self.fixed_ratio:\n+ ow = int(self.short_size * 4.0 / 3.0)\n+ elif not self.keep_ratio: # no\n+ ow = self.short_size\n+ else:\n+ scale_factor = self.short_size / h\n+ oh = int(h * float(scale_factor) +\n+ 0.5) if self.do_round else int(h * self.short_size / w)\n+ ow = int(w * float(scale_factor) +\n+ 0.5) if self.do_round else int(w * self.short_size / h)\n+\n+ if type(img) == np.ndarray:\n+ img = Image.fromarray(img, mode='RGB')\n+\n+ if self.backend == 'pillow':\n+ result_img = img.resize((ow, oh), Image.BILINEAR)\n+ elif self.backend == 'cv2' and (self.keep_ratio is not None):\n+ result_img = cv2.resize(\n+ img, (ow, oh), interpolation=cv2.INTER_LINEAR)\n+ else:\n+ result_img = Image.fromarray(\n+ cv2.resize(\n+ np.asarray(img), (ow, oh), interpolation=cv2.INTER_LINEAR))\n+\n+ return result_img\n+\n+\nclass NormalizeImage(object):\n\"\"\"normalize image\nArgs:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/video_action_infer.py",
"new_path": "deploy/python/video_action_infer.py",
"diff": "@@ -197,7 +197,7 @@ class VideoActionRecognizer(object):\nimg_mean = [0.485, 0.456, 0.406]\nimg_std = [0.229, 0.224, 0.225]\nops = [\n- Scale(self.short_size), CenterCrop(self.target_size), Image2Array(),\n+ CenterCrop(self.target_size), Image2Array(),\nNormalization(img_mean, img_std)\n]\nfor op in ops:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
Scale frames before fight action recognition (#6170)
* Scale frames before fight action recognition
* put
short_size = self.cfg["VIDEO_ACTION"]["short_size"]
scale = Scale(short_size)
out of while
* change class name from Scale to ShortSizeScale
|
499,299 |
14.06.2022 16:10:45
| -28,800 |
51c2ae6e0394e1fc641f5f4c953371dc9c9a3e22
|
fix doc error, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "docs/tutorials/GETTING_STARTED.md",
"new_path": "docs/tutorials/GETTING_STARTED.md",
"diff": "@@ -128,7 +128,7 @@ list below can be viewed by `--help`\n--output_dir=infer_output/ \\\n--draw_threshold=0.5 \\\n-o weights=output/faster_rcnn_r50_fpn_1x_coco/model_final \\\n- --use_vdl=Ture\n+ --use_vdl=True\n```\n`--draw_threshold` is an optional argument. Default is 0.5.\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix doc error, test=document_fix (#6192)
|
499,304 |
15.06.2022 20:47:14
| -28,800 |
941bbf7cb3829e390bc8e1eec3cd76d056b329ac
|
Improve the usability of voc metric
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/metrics/metrics.py",
"new_path": "ppdet/metrics/metrics.py",
"diff": "@@ -248,11 +248,13 @@ class VOCMetric(Metric):\nself.detection_map.reset()\ndef update(self, inputs, outputs):\n- bbox_np = outputs['bbox'].numpy()\n+ bbox_np = outputs['bbox'].numpy() if isinstance(\n+ outputs['bbox'], paddle.Tensor) else outputs['bbox']\nbboxes = bbox_np[:, 2:]\nscores = bbox_np[:, 1]\nlabels = bbox_np[:, 0]\n- bbox_lengths = outputs['bbox_num'].numpy()\n+ bbox_lengths = outputs['bbox_num'].numpy() if isinstance(\n+ outputs['bbox_num'], paddle.Tensor) else outputs['bbox_num']\nif bboxes.shape == (1, 1) or bboxes is None:\nreturn\n@@ -261,18 +263,26 @@ class VOCMetric(Metric):\ndifficults = inputs['difficult'] if not self.evaluate_difficult \\\nelse None\n- scale_factor = inputs['scale_factor'].numpy(\n- ) if 'scale_factor' in inputs else np.ones(\n- (gt_boxes.shape[0], 2)).astype('float32')\n+ if 'scale_factor' in inputs:\n+ scale_factor = inputs['scale_factor'].numpy() if isinstance(\n+ inputs['scale_factor'],\n+ paddle.Tensor) else inputs['scale_factor']\n+ else:\n+ scale_factor = np.ones((gt_boxes.shape[0], 2)).astype('float32')\nbbox_idx = 0\nfor i in range(len(gt_boxes)):\n- gt_box = gt_boxes[i].numpy()\n+ gt_box = gt_boxes[i].numpy() if isinstance(\n+ gt_boxes[i], paddle.Tensor) else gt_boxes[i]\nh, w = scale_factor[i]\ngt_box = gt_box / np.array([w, h, w, h])\n- gt_label = gt_labels[i].numpy()\n- difficult = None if difficults is None \\\n- else difficults[i].numpy()\n+ gt_label = gt_labels[i].numpy() if isinstance(\n+ gt_labels[i], paddle.Tensor) else gt_labels[i]\n+ if difficults is not None:\n+ difficult = difficults[i].numpy() if isinstance(\n+ difficults[i], paddle.Tensor) else difficults[i]\n+ else:\n+ difficult = None\nbbox_num = bbox_lengths[i]\nbbox = bboxes[bbox_idx:bbox_idx + bbox_num]\nscore = scores[bbox_idx:bbox_idx + bbox_num]\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
Improve the usability of voc metric (#6197)
|
499,301 |
16.06.2022 13:52:20
| -28,800 |
145d155623abde6ad9dcc9357c05dfa437ef42a6
|
add layer_norm for convnext outputs
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/convnext.py",
"new_path": "ppdet/modeling/backbones/convnext.py",
"diff": "@@ -141,6 +141,7 @@ class ConvNeXt(nn.Layer):\nlayer_scale_init_value=1e-6,\nhead_init_scale=1.,\nreturn_idx=[1, 2, 3],\n+ norm_output=True,\npretrained=None, ):\nsuper().__init__()\n@@ -178,6 +179,14 @@ class ConvNeXt(nn.Layer):\nself.return_idx = return_idx\nself.dims = [dims[i] for i in return_idx] # [::-1]\n+ self.norm_output = norm_output\n+ if norm_output:\n+ self.norms = nn.LayerList([\n+ LayerNorm(\n+ c, eps=1e-6, data_format=\"channels_first\")\n+ for c in self.dims\n+ ])\n+\nself.apply(self._init_weights)\n# self.head.weight.set_value(self.head.weight.numpy() * head_init_scale)\n# self.head.bias.set_value(self.head.weight.numpy() * head_init_scale)\n@@ -202,9 +211,11 @@ class ConvNeXt(nn.Layer):\nx = self.stages[i](x)\noutput.append(x)\n- output = [output[i] for i in self.return_idx]\n+ outputs = [output[i] for i in self.return_idx]\n+ if self.norm_output:\n+ outputs = [self.norms[i](out) for i, out in enumerate(outputs)]\n- return output\n+ return outputs\ndef forward(self, x):\nx = self.forward_features(x['image'])\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add layer_norm for convnext outputs (#6201)
|
499,339 |
16.06.2022 20:47:21
| -28,800 |
db1b265492311c8c8cf73fe6473486e4c489fd4e
|
[TIPC] fix serving infer shell
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/prepare.sh",
"new_path": "test_tipc/prepare.sh",
"diff": "@@ -80,14 +80,6 @@ elif [ ${MODE} = \"paddle2onnx_infer\" ];then\n${python} -m pip install install paddle2onnx\n${python} -m pip install onnxruntime==1.10.0\nelif [ ${MODE} = \"serving_infer\" ];then\n- git clone https://github.com/PaddlePaddle/Serving\n- cd Serving\n- bash tools/paddle_env_install.sh\n- ${python} -m pip install -r python/requirements.txt\n- cd ..\n- ${python} -m pip install paddle-serving-client -i https://pypi.tuna.tsinghua.edu.cn/simple\n- ${python} -m pip install paddle-serving-app -i https://pypi.tuna.tsinghua.edu.cn/simple\n- ${python} -m pip install paddle-serving-server-gpu -i https://pypi.tuna.tsinghua.edu.cn/simple\nunset https_proxy http_proxy\nelse\n# download coco lite data\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_serving_infer_cpp.sh",
"new_path": "test_tipc/test_serving_infer_cpp.sh",
"diff": "@@ -89,10 +89,6 @@ function func_serving_inference(){\ndone\n}\n-# build paddle_serving_server\n-bash deploy/serving/cpp/build_server.sh\n-echo \"################### build finished! ###################\"\n-\n# run serving infer\nCount=0\nIFS=\"|\"\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_serving_infer_python.sh",
"new_path": "test_tipc/test_serving_infer_python.sh",
"diff": "@@ -81,9 +81,9 @@ function func_serving_inference(){\n}\n# set cuda device\n-GPUID=$2\n+GPUID=$3\nif [ ${#GPUID} -le 0 ];then\n- env=\" \"\n+ env=\"export CUDA_VISIBLE_DEVICES=0\"\nelse\nenv=\"export CUDA_VISIBLE_DEVICES=${GPUID}\"\nfi\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix serving infer shell (#6206)
|
499,301 |
20.06.2022 17:36:36
| -28,800 |
29a5c2fa3e30c07610a41392f1963d2b3d0c0fe5
|
use bias
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/necks/hrfpn.py",
"new_path": "ppdet/modeling/necks/hrfpn.py",
"diff": "@@ -37,7 +37,8 @@ class HRFPN(nn.Layer):\nout_channel=256,\nshare_conv=False,\nextra_stage=1,\n- spatial_scales=[1. / 4, 1. / 8, 1. / 16, 1. / 32]):\n+ spatial_scales=[1. / 4, 1. / 8, 1. / 16, 1. / 32],\n+ use_bias=False):\nsuper(HRFPN, self).__init__()\nin_channel = sum(in_channels)\nself.in_channel = in_channel\n@@ -47,12 +48,14 @@ class HRFPN(nn.Layer):\nspatial_scales = spatial_scales + [spatial_scales[-1] / 2.]\nself.spatial_scales = spatial_scales\nself.num_out = len(self.spatial_scales)\n+ self.use_bias = use_bias\n+ bias_attr = False if use_bias is False else None\nself.reduction = nn.Conv2D(\nin_channels=in_channel,\nout_channels=out_channel,\nkernel_size=1,\n- bias_attr=False)\n+ bias_attr=bias_attr)\nif share_conv:\nself.fpn_conv = nn.Conv2D(\n@@ -60,7 +63,7 @@ class HRFPN(nn.Layer):\nout_channels=out_channel,\nkernel_size=3,\npadding=1,\n- bias_attr=False)\n+ bias_attr=bias_attr)\nelse:\nself.fpn_conv = []\nfor i in range(self.num_out):\n@@ -72,7 +75,7 @@ class HRFPN(nn.Layer):\nout_channels=out_channel,\nkernel_size=3,\npadding=1,\n- bias_attr=False))\n+ bias_attr=bias_attr))\nself.fpn_conv.append(conv)\ndef forward(self, body_feats):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
use bias (#6234)
|
499,392 |
20.06.2022 19:59:30
| -28,800 |
3d45bee1013a3b1b4a7048d67e6842f4c116210c
|
Fix keypoint metric
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/metrics/keypoint_metrics.py",
"new_path": "ppdet/metrics/keypoint_metrics.py",
"diff": "@@ -16,6 +16,7 @@ import os\nimport json\nfrom collections import defaultdict, OrderedDict\nimport numpy as np\n+import paddle\nfrom pycocotools.coco import COCO\nfrom pycocotools.cocoeval import COCOeval\nfrom ..modeling.keypoint_utils import oks_nms\n@@ -70,15 +71,23 @@ class KeyPointTopDownCOCOEval(object):\nself.results['all_preds'][self.idx:self.idx + num_images, :, 0:\n3] = kpts[:, :, 0:3]\nself.results['all_boxes'][self.idx:self.idx + num_images, 0:2] = inputs[\n- 'center'].numpy()[:, 0:2]\n+ 'center'].numpy()[:, 0:2] if isinstance(\n+ inputs['center'], paddle.Tensor) else inputs['center'][:, 0:2]\nself.results['all_boxes'][self.idx:self.idx + num_images, 2:4] = inputs[\n- 'scale'].numpy()[:, 0:2]\n+ 'scale'].numpy()[:, 0:2] if isinstance(\n+ inputs['scale'], paddle.Tensor) else inputs['scale'][:, 0:2]\nself.results['all_boxes'][self.idx:self.idx + num_images, 4] = np.prod(\n- inputs['scale'].numpy() * 200, 1)\n- self.results['all_boxes'][self.idx:self.idx + num_images,\n- 5] = np.squeeze(inputs['score'].numpy())\n+ inputs['scale'].numpy() * 200,\n+ 1) if isinstance(inputs['scale'], paddle.Tensor) else np.prod(\n+ inputs['scale'] * 200, 1)\n+ self.results['all_boxes'][\n+ self.idx:self.idx + num_images,\n+ 5] = np.squeeze(inputs['score'].numpy()) if isinstance(\n+ inputs['score'], paddle.Tensor) else np.squeeze(inputs['score'])\n+ if isinstance(inputs['im_id'], paddle.Tensor):\nself.results['image_path'].extend(inputs['im_id'].numpy())\n-\n+ else:\n+ self.results['image_path'].extend(inputs['im_id'])\nself.idx += num_images\ndef _write_coco_keypoint_results(self, keypoints):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
Fix keypoint metric (#6222)
|
499,301 |
22.06.2022 17:16:48
| -28,800 |
b0fb44b0b97505184140cbc829691e0d7578af3b
|
fix convnext init
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/convnext.py",
"new_path": "ppdet/modeling/backbones/convnext.py",
"diff": "@@ -29,7 +29,7 @@ import numpy as np\nfrom ppdet.core.workspace import register, serializable\nfrom ..shape_spec import ShapeSpec\n-from .transformer_utils import DropPath, trunc_normal_\n+from .transformer_utils import DropPath, trunc_normal_, zeros_\n__all__ = ['ConvNeXt']\n@@ -129,7 +129,6 @@ class ConvNeXt(nn.Layer):\ndims (int): Feature dimension at each stage. Default: [96, 192, 384, 768]\ndrop_path_rate (float): Stochastic depth rate. Default: 0.\nlayer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.\n- head_init_scale (float): Init scaling value for classifier weights and biases. Default: 1.\n\"\"\"\ndef __init__(\n@@ -139,7 +138,6 @@ class ConvNeXt(nn.Layer):\ndims=[96, 192, 384, 768],\ndrop_path_rate=0.,\nlayer_scale_init_value=1e-6,\n- head_init_scale=1.,\nreturn_idx=[1, 2, 3],\nnorm_output=True,\npretrained=None, ):\n@@ -188,8 +186,6 @@ class ConvNeXt(nn.Layer):\n])\nself.apply(self._init_weights)\n- # self.head.weight.set_value(self.head.weight.numpy() * head_init_scale)\n- # self.head.bias.set_value(self.head.weight.numpy() * head_init_scale)\nif pretrained is not None:\nif 'http' in pretrained: #URL\n@@ -201,8 +197,8 @@ class ConvNeXt(nn.Layer):\ndef _init_weights(self, m):\nif isinstance(m, (nn.Conv2D, nn.Linear)):\n- trunc_normal_(m.weight, std=.02)\n- nn.init.constant_(m.bias, 0)\n+ trunc_normal_(m.weight)\n+ zeros_(m.bias)\ndef forward_features(self, x):\noutput = []\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix convnext init (#6248)
|
499,298 |
23.06.2022 12:37:59
| -28,800 |
8d21f781c672889896f0684f22a41d5108825471
|
fix deepsort bytetrack doc
|
[
{
"change_type": "MODIFY",
"old_path": "configs/mot/bytetrack/_base_/mix_det.yml",
"new_path": "configs/mot/bytetrack/_base_/mix_det.yml",
"diff": "@@ -11,7 +11,7 @@ TrainDataset:\nEvalDataset:\n!COCODataSet\n- image_dir: train\n+ image_dir: images/train\nanno_path: annotations/val_half.json\ndataset_dir: dataset/mot/MOT17\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/bytetrack/detector/README_cn.md",
"new_path": "configs/mot/bytetrack/detector/README_cn.md",
"diff": "@@ -30,7 +30,7 @@ job_name=ppyoloe_crn_l_36e_640x640_mot17half\nconfig=configs/mot/bytetrack/detector/${job_name}.yml\nlog_dir=log_dir/${job_name}\n# 1. training\n-python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp --fleet\n+python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp\n# 2. evaluation\nCUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=https://paddledet.bj.bcebos.com/models/mot/${job_name}.pdparams\n```\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/deepsort/deepsort_ppyoloe_pplcnet.yml",
"new_path": "configs/mot/deepsort/deepsort_ppyoloe_pplcnet.yml",
"diff": "@@ -92,7 +92,6 @@ PPYOLOEHead:\ngrid_cell_offset: 0.5\nstatic_assigner_epoch: -1 # 100\nuse_varifocal_loss: True\n- eval_input_size: [640, 640]\nloss_weight: {class: 1.0, iou: 2.5, dfl: 0.5}\nstatic_assigner:\nname: ATSSAssigner\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/deepsort/deepsort_ppyoloe_resnet.yml",
"new_path": "configs/mot/deepsort/deepsort_ppyoloe_resnet.yml",
"diff": "@@ -91,7 +91,6 @@ PPYOLOEHead:\ngrid_cell_offset: 0.5\nstatic_assigner_epoch: -1 # 100\nuse_varifocal_loss: True\n- eval_input_size: [640, 640]\nloss_weight: {class: 1.0, iou: 2.5, dfl: 0.5}\nstatic_assigner:\nname: ATSSAssigner\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/deepsort/detector/ppyoloe_crn_l_36e_640x640_mot17half.yml",
"new_path": "configs/mot/deepsort/detector/ppyoloe_crn_l_36e_640x640_mot17half.yml",
"diff": "@@ -6,6 +6,7 @@ weights: output/ppyoloe_crn_l_36e_640x640_mot17half/model_final\nlog_iter: 20\nsnapshot_epoch: 2\n+\n# schedule configuration for fine-tuning\nepoch: 36\nLearningRate:\n@@ -15,7 +16,7 @@ LearningRate:\nmax_epochs: 43\n- !LinearWarmup\nstart_factor: 0.001\n- steps: 100\n+ epochs: 1\nOptimizerBuilder:\noptimizer:\n@@ -25,9 +26,11 @@ OptimizerBuilder:\nfactor: 0.0005\ntype: L2\n+\nTrainReader:\nbatch_size: 8\n+\n# detector configuration\narchitecture: YOLOv3\nnorm_type: sync_bn\n@@ -62,7 +65,6 @@ PPYOLOEHead:\ngrid_cell_offset: 0.5\nstatic_assigner_epoch: -1 # 100\nuse_varifocal_loss: True\n- eval_input_size: [640, 640]\nloss_weight: {class: 1.0, iou: 2.5, dfl: 0.5}\nstatic_assigner:\nname: ATSSAssigner\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/det_infer.py",
"new_path": "deploy/pptracking/python/det_infer.py",
"diff": "@@ -32,7 +32,7 @@ sys.path.insert(0, parent_path)\nfrom benchmark_utils import PaddleInferBenchmark\nfrom picodet_postprocess import PicoDetPostProcess\n-from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, decode_image\n+from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, Pad, decode_image\nfrom mot.visualize import visualize_box_mask\nfrom mot_utils import argsparser, Timer, get_current_memory_mb\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot_sde_infer.py",
"new_path": "deploy/pptracking/python/mot_sde_infer.py",
"diff": "@@ -186,7 +186,9 @@ class SDE_Detector(Detector):\ndef postprocess(self, inputs, result):\n# postprocess output of predictor\n- np_boxes_num = result['boxes_num']\n+ keep_idx = result['boxes'][:, 1] > self.threshold\n+ result['boxes'] = result['boxes'][keep_idx]\n+ np_boxes_num = [len(result['boxes'])]\nif np_boxes_num[0] <= 0:\nprint('[WARNNING] No object detected.')\nresult = {'boxes': np.zeros([0, 6]), 'boxes_num': [0]}\n@@ -520,8 +522,8 @@ class SDE_Detector(Detector):\n# bs=1 in MOT model\nonline_tlwhs, online_scores, online_ids = mot_results[0]\n- # NOTE: just implement flow statistic for one class\n- if num_classes == 1:\n+ # flow statistic for one class, and only for bytetracker\n+ if num_classes == 1 and not self.use_deepsort_tracker:\nresult = (frame_id + 1, online_tlwhs[0], online_scores[0],\nonline_ids[0])\nstatistic = flow_statistic(\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/preprocess.py",
"new_path": "deploy/pptracking/python/preprocess.py",
"diff": "@@ -245,6 +245,34 @@ class LetterBoxResize(object):\nreturn im, im_info\n+class Pad(object):\n+ def __init__(self, size, fill_value=[114.0, 114.0, 114.0]):\n+ \"\"\"\n+ Pad image to a specified size.\n+ Args:\n+ size (list[int]): image target size\n+ fill_value (list[float]): rgb value of pad area, default (114.0, 114.0, 114.0)\n+ \"\"\"\n+ super(Pad, self).__init__()\n+ if isinstance(size, int):\n+ size = [size, size]\n+ self.size = size\n+ self.fill_value = fill_value\n+\n+ def __call__(self, im, im_info):\n+ im_h, im_w = im.shape[:2]\n+ h, w = self.size\n+ if h == im_h and w == im_w:\n+ im = im.astype(np.float32)\n+ return im, im_info\n+\n+ canvas = np.ones((h, w, 3), dtype=np.float32)\n+ canvas *= np.array(self.fill_value, dtype=np.float32)\n+ canvas[0:im_h, 0:im_w, :] = im.astype(np.float32)\n+ im = canvas\n+ return im, im_info\n+\n+\ndef preprocess(im, preprocess_ops):\n# process image by preprocess_ops\nim_info = {\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix deepsort bytetrack doc (#6255)
|
499,348 |
28.06.2022 18:43:51
| -28,800 |
3428d97fce7e006f247e9a6671edc471da96fdc9
|
rename imshow name;
test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -671,7 +671,7 @@ class PipePredictor(object):\ncenter_traj) # visualize\nwriter.write(im)\nif self.file_name is None: # use camera_id\n- cv2.imshow('PPHuman&&PPVehicle', im)\n+ cv2.imshow('Paddle-Pipeline', im)\nif cv2.waitKey(1) & 0xFF == ord('q'):\nbreak\ncontinue\n@@ -833,7 +833,7 @@ class PipePredictor(object):\ncenter_traj) # visualize\nwriter.write(im)\nif self.file_name is None: # use camera_id\n- cv2.imshow('PPHuman', im)\n+ cv2.imshow('Paddle-Pipeline', im)\nif cv2.waitKey(1) & 0xFF == ord('q'):\nbreak\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
rename imshow name; (#6295)
test=document_fix
|
499,313 |
30.06.2022 14:33:13
| -28,800 |
d2e7bd38c30b23fea75f67662ee8e4a5eb415bd0
|
fix spine_coco.tar md5sum
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/utils/download.py",
"new_path": "ppdet/utils/download.py",
"diff": "@@ -97,7 +97,7 @@ DATASETS = {\n'49ce5a9b5ad0d6266163cd01de4b018e', ), ], ['annotations', 'images']),\n'spine_coco': ([(\n'https://paddledet.bj.bcebos.com/data/spine_coco.tar',\n- '7ed69ae73f842cd2a8cf4f58dc3c5535', ), ], ['annotations', 'images']),\n+ '03030f42d9b6202a6e425d4becefda0d', ), ], ['annotations', 'images']),\n'mot': (),\n'objects365': (),\n'coco_ce': ([(\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix spine_coco.tar md5sum (#6309)
|
499,298 |
30.06.2022 18:05:12
| -28,800 |
00b656f2f9d84e2cbe5689cd9ce29273b8466c6c
|
update ppyoloe test AP in all docs
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"diff": "@@ -39,7 +39,7 @@ PPYOLOEHead:\nbeta: 6.0\nnms:\nname: MultiClassNMS\n- nms_top_k: 1000\n- keep_top_k: 100\n+ nms_top_k: 10000\n+ keep_top_k: 300\nscore_threshold: 0.01\n- nms_threshold: 0.6\n+ nms_threshold: 0.7\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update ppyoloe test AP in all docs (#6315)
|
499,339 |
01.07.2022 11:28:15
| -28,800 |
bccee30e34c4e8f86672b4e4482ece03daa96ecc
|
[TIPC] fix ptq txt, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/keypoint/tinypose_128x96_train_ptq_infer_python.txt",
"new_path": "test_tipc/configs/keypoint/tinypose_128x96_train_ptq_infer_python.txt",
"diff": "@@ -5,7 +5,7 @@ filename:\n##\n--output_dir:./output_inference\nweights:https://paddledet.bj.bcebos.com/models/keypoint/tinypose_128x96.pdparams\n-kl_quant_export:tools/post_quant.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+kl_quant_export:tools/post_quant.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config configs/slim/post_quant/tinypose_128x96_ptq.yml -o\nexport_param1:null\n##\ninference:./deploy/python/keypoint_infer.py\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_pact_infer_python.txt",
"new_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_pact_infer_python.txt",
"diff": "@@ -14,7 +14,7 @@ filename:null\n##\ntrainer:pact_train\nnorm_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml -o\n-pact_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config configs/slim/quant/yolov3_mobilenet_v3_qat.yml -o\n+pact_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config configs/slim/quant/mask_rcnn_r50_fpn_1x_qat.yml -o\nfpgm_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config _template_fpgm -o\ndistill_train:null\nnull:null\n@@ -41,7 +41,7 @@ inference:./deploy/python/infer.py\n--device:gpu|cpu\n--enable_mkldnn:False\n--cpu_threads:4\n---batch_size:1|2\n+--batch_size:1\n--use_tensorrt:null\n--run_mode:paddle\n--model_dir:\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_ptq_infer_python.txt",
"new_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_ptq_infer_python.txt",
"diff": "@@ -5,14 +5,14 @@ filename:\n##\n--output_dir:./output_inference\nweights:https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_1x_coco.pdparams\n-kl_quant_export:tools/post_quant.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+kl_quant_export:tools/post_quant.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config configs/slim/post_quant/mask_rcnn_r50_fpn_1x_coco_ptq.yml -o\nexport_param1:null\n##\ninference:./deploy/python/infer.py\n--device:gpu|cpu\n--enable_mkldnn:False\n--cpu_threads:4\n---batch_size:1|2\n+--batch_size:1\n--run_mode:paddle\n--model_dir:\n--image_dir:./dataset/coco/test2017/\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyolo/ppyolo_mbv3_large_coco_train_pact_infer_python.txt",
"new_path": "test_tipc/configs/ppyolo/ppyolo_mbv3_large_coco_train_pact_infer_python.txt",
"diff": "@@ -14,7 +14,7 @@ filename:null\n##\ntrainer:pact_train\nnorm_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml -o\n-pact_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/quant/yolov3_mobilenet_v3_qat.yml -o\n+pact_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/quant/ppyolo_mbv3_large_qat.yml -o\nfpgm_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/prune/ppyolo_mbv3_large_prune_fpgm.yml -o\ndistill_train:null\nnull:null\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyolo/ppyolo_mbv3_large_coco_train_ptq_infer_python.txt",
"new_path": "test_tipc/configs/ppyolo/ppyolo_mbv3_large_coco_train_ptq_infer_python.txt",
"diff": "@@ -5,7 +5,7 @@ filename:\n##\n--output_dir:./output_inference\nweights:https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams\n-kl_quant_export:tools/post_quant.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+kl_quant_export:tools/post_quant.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/post_quant/ppyolo_mbv3_large_ptq.yml -o\nexport_param1:null\n##\ninference:./deploy/python/infer.py\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyoloe/ppyoloe_crn_s_300e_coco_train_ptq_infer_python.txt",
"new_path": "test_tipc/configs/ppyoloe/ppyoloe_crn_s_300e_coco_train_ptq_infer_python.txt",
"diff": "@@ -5,7 +5,7 @@ filename:\n##\n--output_dir:./output_inference\nweights:https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_300e_coco.pdparams\n-kl_quant_export:tools/post_quant.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+kl_quant_export:tools/post_quant.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml --slim_config configs/slim/post_quant/ppyoloe_crn_s_300e_coco_ptq.yml -o\nexport_param1:null\n##\ninference:./deploy/python/infer.py\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix ptq txt, test=document_fix (#6320)
|
499,354 |
04.07.2022 18:32:27
| -28,800 |
1f13295326a8d7584976c69c92d7c68f1493c208
|
mobileone block k>1 bugfix
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/mobileone.py",
"new_path": "ppdet/modeling/backbones/mobileone.py",
"diff": "@@ -22,7 +22,7 @@ import paddle\nimport paddle.nn as nn\nfrom paddle import ParamAttr\nfrom paddle.regularizer import L2Decay\n-from paddle.nn.initializer import Normal\n+from paddle.nn.initializer import Normal, Constant\nfrom ppdet.modeling.ops import get_act_fn\nfrom ppdet.modeling.layers import ConvNormLayer\n@@ -57,9 +57,7 @@ class MobileOneBlock(nn.Layer):\nself.depth_conv = nn.LayerList()\nself.point_conv = nn.LayerList()\n- for i in range(self.k):\n- if i > 0:\n- stride = 1\n+ for _ in range(self.k):\nself.depth_conv.append(\nConvNormLayer(\nch_in,\n@@ -112,7 +110,8 @@ class MobileOneBlock(nn.Layer):\nself.rbr_identity_st2 = nn.BatchNorm2D(\nnum_features=ch_out,\nweight_attr=ParamAttr(regularizer=L2Decay(0.0)),\n- bias_attr=ParamAttr(regularizer=L2Decay(0.0)))\n+ bias_attr=ParamAttr(regularizer=L2Decay(\n+ 0.0))) if ch_in == ch_out and self.stride == 1 else None\nself.act = get_act_fn(act) if act is None or isinstance(act, (\nstr, dict)) else act\n@@ -125,9 +124,10 @@ class MobileOneBlock(nn.Layer):\nelse:\nid_out_st1 = self.rbr_identity_st1(x)\n- x1_1 = x.clone()\n+ x1_1 = 0\nfor i in range(self.k):\n- x1_1 = self.depth_conv[i](x1_1)\n+ x1_1 += self.depth_conv[i](x)\n+\nx1_2 = self.rbr_1x1(x)\nx1 = self.act(x1_1 + x1_2 + id_out_st1)\n@@ -136,9 +136,9 @@ class MobileOneBlock(nn.Layer):\nelse:\nid_out_st2 = self.rbr_identity_st2(x1)\n- x2_1 = x1.clone()\n+ x2_1 = 0\nfor i in range(self.k):\n- x2_1 = self.point_conv[i](x2_1)\n+ x2_1 += self.point_conv[i](x1)\ny = self.act(x2_1 + id_out_st2)\nreturn y\n@@ -151,7 +151,9 @@ class MobileOneBlock(nn.Layer):\nkernel_size=self.kernel_size,\nstride=self.stride,\npadding=self.padding,\n- groups=self.ch_in)\n+ groups=self.ch_in,\n+ bias_attr=ParamAttr(\n+ initializer=Constant(value=0.), learning_rate=1.))\nif not hasattr(self, 'conv2'):\nself.conv2 = nn.Conv2D(\nin_channels=self.ch_in,\n@@ -159,7 +161,9 @@ class MobileOneBlock(nn.Layer):\nkernel_size=1,\nstride=1,\npadding='SAME',\n- groups=1)\n+ groups=1,\n+ bias_attr=ParamAttr(\n+ initializer=Constant(value=0.), learning_rate=1.))\nconv1_kernel, conv1_bias, conv2_kernel, conv2_bias = self.get_equivalent_kernel_bias(\n)\n@@ -211,26 +215,24 @@ class MobileOneBlock(nn.Layer):\nreturn 0, 0\nif isinstance(branch, nn.LayerList):\n- kernel = 0\n- running_mean = 0\n- running_var = 0\n- gamma = 0\n- beta = 0\n- eps = 0\n+ fused_kernels = []\n+ fused_bias = []\nfor block in branch:\n- kernel += block.conv.weight\n- running_mean += block.norm._mean\n- running_var += block.norm._variance\n- gamma += block.norm.weight\n- beta += block.norm.bias\n- eps += block.norm._epsilon\n+ kernel = block.conv.weight\n+ running_mean = block.norm._mean\n+ running_var = block.norm._variance\n+ gamma = block.norm.weight\n+ beta = block.norm.bias\n+ eps = block.norm._epsilon\n+\n+ std = (running_var + eps).sqrt()\n+ t = (gamma / std).reshape((-1, 1, 1, 1))\n+\n+ fused_kernels.append(kernel * t)\n+ fused_bias.append(beta - running_mean * gamma / std)\n+\n+ return sum(fused_kernels), sum(fused_bias)\n- kernel /= len(branch)\n- running_mean /= len(branch)\n- running_var /= len(branch)\n- gamma /= len(branch)\n- beta /= len(branch)\n- eps /= len(branch)\nelif isinstance(branch, ConvNormLayer):\nkernel = branch.conv.weight\nrunning_mean = branch.norm._mean\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
mobileone block k>1 bugfix (#6342)
|
499,348 |
04.07.2022 19:38:44
| -28,800 |
fe9e983daeeb49be9a4af38b80cc4967afc1690f
|
add vehicle times
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipe_utils.py",
"new_path": "deploy/pipeline/pipe_utils.py",
"diff": "@@ -157,7 +157,8 @@ class PipeTimer(Times):\n'reid': Times(),\n'det_action': Times(),\n'cls_action': Times(),\n- 'vehicle_attr': Times()\n+ 'vehicle_attr': Times(),\n+ 'vehicleplate': Times()\n}\nself.img_num = 0\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipeline.py",
"new_path": "deploy/pipeline/pipeline.py",
"diff": "@@ -679,8 +679,12 @@ class PipePredictor(object):\nframe_rgb, mot_res)\nif self.with_vehicleplate:\n+ if frame_id > self.warmup_frame:\n+ self.pipe_timer.module_time['vehicleplate'].start()\nplatelicense = self.vehicleplate_detector.get_platelicense(\ncrop_input)\n+ if frame_id > self.warmup_frame:\n+ self.pipe_timer.module_time['vehicleplate'].end()\nself.pipeline_res.update(platelicense, 'vehicleplate')\nif self.with_human_attr:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add vehicle times (#6327)
|
499,363 |
04.07.2022 19:47:17
| -28,800 |
95e07d186bf940f087d7672e45de973f16da1048
|
Develop: get fight recognition model from model dir
* Update action.md
delete the step of change model name of fight recognition
* Update video_action_infer.py
get model path from model_dir without model name
* optimize action vis
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipeline.py",
"new_path": "deploy/pipeline/pipeline.py",
"diff": "@@ -919,9 +919,12 @@ class PipePredictor(object):\nvideo_action_score = None\nif video_action_res and video_action_res[\"class\"] == 1:\nvideo_action_score = video_action_res[\"score\"]\n+ mot_boxes = None\n+ if mot_res:\n+ mot_boxes = mot_res['boxes']\nimage = visualize_action(\nimage,\n- mot_res['boxes'],\n+ mot_boxes,\naction_visual_collector=None,\naction_text=\"SkeletonAction\",\nvideo_action_score=video_action_score,\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pphuman/video_action_infer.py",
"new_path": "deploy/pipeline/pphuman/video_action_infer.py",
"diff": "@@ -96,8 +96,8 @@ class VideoActionRecognizer(object):\nself.recognize_times = Timer()\n- model_file_path = os.path.join(model_dir, \"ppTSM.pdmodel\")\n- params_file_path = os.path.join(model_dir, \"ppTSM.pdiparams\")\n+ model_file_path = glob.glob(os.path.join(model_dir, \"*.pdmodel\"))[0]\n+ params_file_path = glob.glob(os.path.join(model_dir, \"*.pdiparams\"))[0]\nself.config = Config(model_file_path, params_file_path)\nif device == \"GPU\" or device == \"gpu\":\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
Develop: get fight recognition model from model dir (#6324)
* Update action.md
delete the step of change model name of fight recognition
* Update video_action_infer.py
get model path from model_dir without model name
* optimize action vis
|
499,354 |
05.07.2022 10:58:32
| -28,800 |
dcadfc3e8637d272113811fbf6026633eac17e5f
|
mobileone block bugfix
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/mobileone.py",
"new_path": "ppdet/modeling/backbones/mobileone.py",
"diff": "@@ -178,8 +178,6 @@ class MobileOneBlock(nn.Layer):\nself.__delattr__('rbr_identity_st1')\nif hasattr(self, 'rbr_identity_st2'):\nself.__delattr__('rbr_identity_st2')\n- if hasattr(self, 'id_tensor'):\n- self.__delattr__('id_tensor')\ndef get_equivalent_kernel_bias(self):\nst1_kernel3x3, st1_bias3x3 = self._fuse_bn_tensor(self.depth_conv)\n@@ -248,7 +246,8 @@ class MobileOneBlock(nn.Layer):\ndtype='float32')\nif kernel_size > 1:\nfor i in range(self.ch_in):\n- kernel_value[i, i % input_dim, 1, 1] = 1\n+ kernel_value[i, i % input_dim, (kernel_size - 1) // 2, (\n+ kernel_size - 1) // 2] = 1\nelif kernel_size == 1:\nfor i in range(self.ch_in):\nkernel_value[i, i % input_dim, 0, 0] = 1\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
mobileone block bugfix (#6358)
|
499,299 |
08.07.2022 10:43:51
| -28,800 |
23e9cb95f95c9e42aa4f40c68789da417ddf6d72
|
add frame-skip to boost inference
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/config/infer_cfg_pphuman.yml",
"new_path": "deploy/pipeline/config/infer_cfg_pphuman.yml",
"diff": "@@ -50,6 +50,7 @@ ID_BASED_DETACTION:\nbasemode: \"idbased\"\nthreshold: 0.6\ndisplay_frames: 80\n+ skip_frame_num: 2\nenable: False\nID_BASED_CLSACTION:\n@@ -58,6 +59,7 @@ ID_BASED_CLSACTION:\nbasemode: \"idbased\"\nthreshold: 0.8\ndisplay_frames: 80\n+ skip_frame_num: 2\nenable: False\nREID:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipeline.py",
"new_path": "deploy/pipeline/pipeline.py",
"diff": "@@ -342,7 +342,9 @@ class PipePredictor(object):\nbasemode = idbased_detaction_cfg['basemode']\nthreshold = idbased_detaction_cfg['threshold']\ndisplay_frames = idbased_detaction_cfg['display_frames']\n+ skip_frame_num = idbased_detaction_cfg['skip_frame_num']\nself.modebase[basemode] = True\n+\nself.det_action_predictor = DetActionRecognizer(\nmodel_dir,\ndevice,\n@@ -355,7 +357,8 @@ class PipePredictor(object):\ncpu_threads,\nenable_mkldnn,\nthreshold=threshold,\n- display_frames=display_frames)\n+ display_frames=display_frames,\n+ skip_frame_num=skip_frame_num)\nself.det_action_visual_helper = ActionVisualHelper(1)\nif self.with_idbased_clsaction:\n@@ -366,6 +369,8 @@ class PipePredictor(object):\nthreshold = idbased_clsaction_cfg['threshold']\nself.modebase[basemode] = True\ndisplay_frames = idbased_clsaction_cfg['display_frames']\n+ skip_frame_num = idbased_clsaction_cfg['skip_frame_num']\n+\nself.cls_action_predictor = ClsActionRecognizer(\nmodel_dir,\ndevice,\n@@ -378,7 +383,8 @@ class PipePredictor(object):\ncpu_threads,\nenable_mkldnn,\nthreshold=threshold,\n- display_frames=display_frames)\n+ display_frames=display_frames,\n+ skip_frame_num=skip_frame_num)\nself.cls_action_visual_helper = ActionVisualHelper(1)\nif self.with_skeleton_action:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pphuman/action_infer.py",
"new_path": "deploy/pipeline/pphuman/action_infer.py",
"diff": "@@ -280,6 +280,10 @@ class DetActionRecognizer(object):\nenable_mkldnn (bool): whether to open MKLDNN\nthreshold (float): The threshold of score for action feature object detection.\ndisplay_frames (int): The duration for corresponding detected action.\n+ skip_frame_num (int): The number of frames for interval prediction. A skipped frame will\n+ reuse the result of its last frame. If it is set to 0, no frame will be skipped. Default\n+ is 0.\n+\n\"\"\"\ndef __init__(self,\n@@ -295,7 +299,8 @@ class DetActionRecognizer(object):\nenable_mkldnn=False,\noutput_dir='output',\nthreshold=0.5,\n- display_frames=20):\n+ display_frames=20,\n+ skip_frame_num=0):\nsuper(DetActionRecognizer, self).__init__()\nself.detector = Detector(\nmodel_dir=model_dir,\n@@ -313,10 +318,21 @@ class DetActionRecognizer(object):\nself.threshold = threshold\nself.frame_life = display_frames\nself.result_history = {}\n+ self.skip_frame_num = skip_frame_num\n+ self.skip_frame_cnt = 0\n+ self.id_in_last_frame = []\ndef predict(self, images, mot_result):\n+ if self.skip_frame_cnt == 0 or (not self.check_id_is_same(mot_result)):\ndet_result = self.detector.predict_image(images, visual=False)\nresult = self.postprocess(det_result, mot_result)\n+ else:\n+ result = self.reuse_result(mot_result)\n+\n+ self.skip_frame_cnt += 1\n+ if self.skip_frame_cnt >= self.skip_frame_num:\n+ self.skip_frame_cnt = 0\n+\nreturn result\ndef postprocess(self, det_result, mot_result):\n@@ -343,10 +359,11 @@ class DetActionRecognizer(object):\nif valid_boxes.shape[0] >= 1:\naction_ret['class'] = valid_boxes[0, 0]\naction_ret['score'] = valid_boxes[0, 1]\n- self.result_history[tracker_id] = [0, self.frame_life]\n+ self.result_history[\n+ tracker_id] = [0, self.frame_life, valid_boxes[0, 1]]\nelse:\n- history_det, life_remain = self.result_history.get(tracker_id,\n- [1, 0])\n+ history_det, life_remain, history_score = self.result_history.get(\n+ tracker_id, [1, self.frame_life, -1.0])\naction_ret['class'] = history_det\naction_ret['score'] = -1.0\nlife_remain -= 1\n@@ -354,10 +371,48 @@ class DetActionRecognizer(object):\ndel (self.result_history[tracker_id])\nelif tracker_id in self.result_history:\nself.result_history[tracker_id][1] = life_remain\n+ else:\n+ self.result_history[tracker_id] = [\n+ history_det, life_remain, history_score\n+ ]\nmot_id.append(tracker_id)\nact_res.append(action_ret)\nresult = list(zip(mot_id, act_res))\n+ self.id_in_last_frame = mot_id\n+\n+ return result\n+\n+ def check_id_is_same(self, mot_result):\n+ mot_bboxes = mot_result.get('boxes')\n+ for idx in range(len(mot_bboxes)):\n+ tracker_id = mot_bboxes[idx, 0]\n+ if tracker_id not in self.id_in_last_frame:\n+ return False\n+ return True\n+\n+ def reuse_result(self, mot_result):\n+ # This function reusing previous results of the same ID directly.\n+ mot_bboxes = mot_result.get('boxes')\n+\n+ mot_id = []\n+ act_res = []\n+\n+ for idx in range(len(mot_bboxes)):\n+ tracker_id = mot_bboxes[idx, 0]\n+ history_cls, life_remain, history_score = self.result_history.get(\n+ tracker_id, [1, 0, -1.0])\n+\n+ life_remain -= 1\n+ if tracker_id in self.result_history:\n+ self.result_history[tracker_id][1] = life_remain\n+\n+ action_ret = {'class': history_cls, 'score': history_score}\n+ mot_id.append(tracker_id)\n+ act_res.append(action_ret)\n+\n+ result = list(zip(mot_id, act_res))\n+ self.id_in_last_frame = mot_id\nreturn result\n@@ -378,6 +433,9 @@ class ClsActionRecognizer(AttrDetector):\nenable_mkldnn (bool): whether to open MKLDNN\nthreshold (float): The threshold of score for action feature object detection.\ndisplay_frames (int): The duration for corresponding detected action.\n+ skip_frame_num (int): The number of frames for interval prediction. A skipped frame will\n+ reuse the result of its last frame. If it is set to 0, no frame will be skipped. Default\n+ is 0.\n\"\"\"\ndef __init__(self,\n@@ -393,7 +451,8 @@ class ClsActionRecognizer(AttrDetector):\nenable_mkldnn=False,\noutput_dir='output',\nthreshold=0.5,\n- display_frames=80):\n+ display_frames=80,\n+ skip_frame_num=0):\nsuper(ClsActionRecognizer, self).__init__(\nmodel_dir=model_dir,\ndevice=device,\n@@ -410,11 +469,22 @@ class ClsActionRecognizer(AttrDetector):\nself.threshold = threshold\nself.frame_life = display_frames\nself.result_history = {}\n+ self.skip_frame_num = skip_frame_num\n+ self.skip_frame_cnt = 0\n+ self.id_in_last_frame = []\ndef predict_with_mot(self, images, mot_result):\n+ if self.skip_frame_cnt == 0 or (not self.check_id_is_same(mot_result)):\nimages = self.crop_half_body(images)\ncls_result = self.predict_image(images, visual=False)[\"output\"]\nresult = self.match_action_with_id(cls_result, mot_result)\n+ else:\n+ result = self.reuse_result(mot_result)\n+\n+ self.skip_frame_cnt += 1\n+ if self.skip_frame_cnt >= self.skip_frame_num:\n+ self.skip_frame_cnt = 0\n+\nreturn result\ndef crop_half_body(self, images):\n@@ -456,8 +526,8 @@ class ClsActionRecognizer(AttrDetector):\n# Current now, class 0 is positive, class 1 is negative.\nif cls_id_res == 1 or (cls_id_res == 0 and\ncls_score_res < self.threshold):\n- history_cls, life_remain = self.result_history.get(tracker_id,\n- [1, 0])\n+ history_cls, life_remain, history_score = self.result_history.get(\n+ tracker_id, [1, self.frame_life, -1.0])\ncls_id_res = history_cls\ncls_score_res = 1 - cls_score_res\nlife_remain -= 1\n@@ -466,12 +536,50 @@ class ClsActionRecognizer(AttrDetector):\nelif tracker_id in self.result_history:\nself.result_history[tracker_id][1] = life_remain\nelse:\n- self.result_history[tracker_id] = [cls_id_res, self.frame_life]\n+ self.result_history[\n+ tracker_id] = [cls_id_res, life_remain, cls_score_res]\n+ else:\n+ self.result_history[\n+ tracker_id] = [cls_id_res, self.frame_life, cls_score_res]\naction_ret = {'class': cls_id_res, 'score': cls_score_res}\nmot_id.append(tracker_id)\nact_res.append(action_ret)\nresult = list(zip(mot_id, act_res))\n+ self.id_in_last_frame = mot_id\n+\n+ return result\n+\n+ def check_id_is_same(self, mot_result):\n+ mot_bboxes = mot_result.get('boxes')\n+ for idx in range(len(mot_bboxes)):\n+ tracker_id = mot_bboxes[idx, 0]\n+ if tracker_id not in self.id_in_last_frame:\n+ return False\n+ return True\n+\n+ def reuse_result(self, mot_result):\n+ # This function reusing previous results of the same ID directly.\n+ mot_bboxes = mot_result.get('boxes')\n+\n+ mot_id = []\n+ act_res = []\n+\n+ for idx in range(len(mot_bboxes)):\n+ tracker_id = mot_bboxes[idx, 0]\n+ history_cls, life_remain, history_score = self.result_history.get(\n+ tracker_id, [1, 0, -1.0])\n+\n+ life_remain -= 1\n+ if tracker_id in self.result_history:\n+ self.result_history[tracker_id][1] = life_remain\n+\n+ action_ret = {'class': history_cls, 'score': history_score}\n+ mot_id.append(tracker_id)\n+ act_res.append(action_ret)\n+\n+ result = list(zip(mot_id, act_res))\n+ self.id_in_last_frame = mot_id\nreturn result\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add frame-skip to boost inference (#6383)
|
499,339 |
08.07.2022 10:58:01
| -28,800 |
f0b524486725d98bb3ba18b2c7ae34c3254d516a
|
[TIPC] add tinypose_KL txt, test=document_fix
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/keypoint/tinypose_128x96_KL_model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt",
"diff": "+===========================cpp_infer_params===========================\n+model_name:tinypose_128x96_KL\n+python:python3.7\n+filename:null\n+##\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/keypoint/tinypose_128x96.pdparams\n+norm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml -o\n+quant_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_kl_quant -o\n+##\n+opencv_dir:default\n+infer_mode:null\n+infer_quant:True\n+inference:./deploy/cpp/build/main\n+--device:gpu|cpu\n+--use_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir_keypoint:\n+--image_dir:./dataset/coco/test2017/\n+--run_benchmark:False\n+--model_dir:./output_inference/picodet_s_320_pedestrian\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/keypoint/tinypose_128x96_KL_model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt",
"diff": "+===========================serving_infer_cpp_params===========================\n+model_name:tinypose_128x96_KL\n+python:python3.7\n+filename:null\n+##\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/keypoint/tinypose_128x96.pdparams\n+norm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --export_serving_model True -o\n+quant_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_pact --export_serving_model True -o\n+fpgm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_fpgm --export_serving_model True -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config configs/slim/post_quant/tinypose_128x96_ptq.yml --export_serving_model True -o\n+##\n+infer_mode:null\n+infer_quant:True\n+--model:null\n+--op:tinypose_128x96\n+--port:9997\n+--gpu_ids:null|0\n+null:null\n+http_client:deploy/serving/cpp/serving_client.py\n+--serving_client:null\n+--image_file:./demo/hrnet_demo.jpg\n+null:null\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/keypoint/tinypose_128x96_KL_model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt",
"diff": "+===========================serving_infer_python_params===========================\n+model_name:tinypose_128x96_KL\n+python:python3.7\n+filename:null\n+##\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/keypoint/tinypose_128x96.pdparams\n+norm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --export_serving_model True -o\n+quant_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_pact --export_serving_model True -o\n+fpgm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_fpgm --export_serving_model True -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config configs/slim/post_quant/tinypose_128x96_ptq.yml --export_serving_model True -o\n+##\n+infer_mode:null\n+infer_quant:True\n+web_service:deploy/serving/python/web_service.py --config=deploy/serving/python/config.yml\n+--model_dir:null\n+--opt:cpu:op.ppdet.local_service_conf.device_type=0|gpu:op.ppdet.local_service_conf.device_type=1\n+null:null\n+http_client:deploy/serving/python/pipeline_http_client.py\n+--image_file:./demo/hrnet_demo.jpg\n+null:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/prepare.sh",
"new_path": "test_tipc/prepare.sh",
"diff": "@@ -58,6 +58,10 @@ elif [ ${MODE} = \"cpp_infer\" ];then\nwget -nc -P ./output_inference/mask_rcnn_r50_fpn_1x_coco_KL/ https://bj.bcebos.com/v1/paddledet/data/tipc/models/mask_rcnn_r50_fpn_1x_coco_ptq.tar --no-check-certificate\ncd ./output_inference/mask_rcnn_r50_fpn_1x_coco_KL/ && tar -xvf mask_rcnn_r50_fpn_1x_coco_ptq.tar && mv -n mask_rcnn_r50_fpn_1x_coco_ptq/* .\ncd ../../\n+ elif [[ ${model_name} = \"tinypose_128x96_KL\" ]]; then\n+ wget -nc -P ./output_inference/tinypose_128x96_KL/ https://bj.bcebos.com/v1/paddledet/data/tipc/models/tinypose_128x96_ptq.tar --no-check-certificate\n+ cd ./output_inference/tinypose_128x96_KL/ && tar -xvf tinypose_128x96_ptq.tar && mv -n tinypose_128x96_ptq/* .\n+ cd ../../\nfi\n# download mot lite data\nwget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/tipc/mot_tipc.tar --no-check-certificate\n@@ -124,6 +128,11 @@ elif [ ${MODE} = \"serving_infer\" ];then\ncd ./output_inference/mask_rcnn_r50_fpn_1x_coco_KL/ && tar -xvf mask_rcnn_r50_fpn_1x_coco_ptq.tar && mv -n mask_rcnn_r50_fpn_1x_coco_ptq/* .\ncd ../../\neval \"${python} -m paddle_serving_client.convert --dirname output_inference/mask_rcnn_r50_fpn_1x_coco_KL/ --model_filename model.pdmodel --params_filename model.pdiparams --serving_server output_inference/mask_rcnn_r50_fpn_1x_coco_KL/serving_server --serving_client output_inference/mask_rcnn_r50_fpn_1x_coco_KL/serving_client\"\n+ elif [[ ${model_name} = \"tinypose_128x96_KL\" ]]; then\n+ wget -nc -P ./output_inference/tinypose_128x96_KL/ https://bj.bcebos.com/v1/paddledet/data/tipc/models/tinypose_128x96_ptq.tar --no-check-certificate\n+ cd ./output_inference/tinypose_128x96_KL/ && tar -xvf tinypose_128x96_ptq.tar && mv -n tinypose_128x96_ptq/* .\n+ cd ../../\n+ eval \"${python} -m paddle_serving_client.convert --dirname output_inference/tinypose_128x96_KL/ --model_filename model.pdmodel --params_filename model.pdiparams --serving_server output_inference/tinypose_128x96_KL/serving_server --serving_client output_inference/tinypose_128x96_KL/serving_client\"\nfi\nelse\n# download coco lite data\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] add tinypose_KL txt, test=document_fix (#6384)
|
499,301 |
11.07.2022 16:42:48
| -28,800 |
e6d4d2bc7ba5eb4aa543e3439fa4e24cdd68d028
|
fix export_model for swin
|
[
{
"change_type": "MODIFY",
"old_path": "configs/faster_rcnn/_base_/faster_rcnn_swin_reader.yml",
"new_path": "configs/faster_rcnn/_base_/faster_rcnn_swin_reader.yml",
"diff": "@@ -30,14 +30,12 @@ EvalReader:\nTestReader:\ninputs_def:\n- image_shape: [1, 3, 640, 640]\n+ image_shape: [-1, 3, 640, 640]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [640, 640], keep_ratio: True}\n+ - LetterBoxResize: {target_size: 640}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\n- batch_transforms:\n- - PadBatch: {pad_to_stride: 32}\nbatch_size: 1\nshuffle: false\ndrop_last: false\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/swin_transformer.py",
"new_path": "ppdet/modeling/backbones/swin_transformer.py",
"diff": "@@ -20,7 +20,6 @@ MIT License [see LICENSE for details]\nimport paddle\nimport paddle.nn as nn\nimport paddle.nn.functional as F\n-from paddle.nn.initializer import TruncatedNormal, Constant, Assign\nfrom ppdet.modeling.shape_spec import ShapeSpec\nfrom ppdet.core.workspace import register, serializable\nimport numpy as np\n@@ -64,7 +63,7 @@ def window_partition(x, window_size):\n\"\"\"\nB, H, W, C = x.shape\nx = x.reshape(\n- [B, H // window_size, window_size, W // window_size, window_size, C])\n+ [-1, H // window_size, window_size, W // window_size, window_size, C])\nwindows = x.transpose([0, 1, 3, 2, 4, 5]).reshape(\n[-1, window_size, window_size, C])\nreturn windows\n@@ -80,10 +79,11 @@ def window_reverse(windows, window_size, H, W):\nReturns:\nx: (B, H, W, C)\n\"\"\"\n+ _, _, _, C = windows.shape\nB = int(windows.shape[0] / (H * W / window_size / window_size))\nx = windows.reshape(\n- [B, H // window_size, W // window_size, window_size, window_size, -1])\n- x = x.transpose([0, 1, 3, 2, 4, 5]).reshape([B, H, W, -1])\n+ [-1, H // window_size, W // window_size, window_size, window_size, C])\n+ x = x.transpose([0, 1, 3, 2, 4, 5]).reshape([-1, H, W, C])\nreturn x\n@@ -158,14 +158,14 @@ class WindowAttention(nn.Layer):\n\"\"\"\nB_, N, C = x.shape\nqkv = self.qkv(x).reshape(\n- [B_, N, 3, self.num_heads, C // self.num_heads]).transpose(\n+ [-1, N, 3, self.num_heads, C // self.num_heads]).transpose(\n[2, 0, 3, 1, 4])\nq, k, v = qkv[0], qkv[1], qkv[2]\nq = q * self.scale\nattn = paddle.mm(q, k.transpose([0, 1, 3, 2]))\n- index = self.relative_position_index.reshape([-1])\n+ index = self.relative_position_index.flatten()\nrelative_position_bias = paddle.index_select(\nself.relative_position_bias_table, index)\n@@ -179,7 +179,7 @@ class WindowAttention(nn.Layer):\nif mask is not None:\nnW = mask.shape[0]\n- attn = attn.reshape([B_ // nW, nW, self.num_heads, N, N\n+ attn = attn.reshape([-1, nW, self.num_heads, N, N\n]) + mask.unsqueeze(1).unsqueeze(0)\nattn = attn.reshape([-1, self.num_heads, N, N])\nattn = self.softmax(attn)\n@@ -189,7 +189,7 @@ class WindowAttention(nn.Layer):\nattn = self.attn_drop(attn)\n# x = (attn @ v).transpose(1, 2).reshape([B_, N, C])\n- x = paddle.mm(attn, v).transpose([0, 2, 1, 3]).reshape([B_, N, C])\n+ x = paddle.mm(attn, v).transpose([0, 2, 1, 3]).reshape([-1, N, C])\nx = self.proj(x)\nx = self.proj_drop(x)\nreturn x\n@@ -267,7 +267,7 @@ class SwinTransformerBlock(nn.Layer):\nshortcut = x\nx = self.norm1(x)\n- x = x.reshape([B, H, W, C])\n+ x = x.reshape([-1, H, W, C])\n# pad feature maps to multiples of window size\npad_l = pad_t = 0\n@@ -289,7 +289,7 @@ class SwinTransformerBlock(nn.Layer):\nx_windows = window_partition(\nshifted_x, self.window_size) # nW*B, window_size, window_size, C\nx_windows = x_windows.reshape(\n- [-1, self.window_size * self.window_size,\n+ [x_windows.shape[0], self.window_size * self.window_size,\nC]) # nW*B, window_size*window_size, C\n# W-MSA/SW-MSA\n@@ -298,7 +298,7 @@ class SwinTransformerBlock(nn.Layer):\n# merge windows\nattn_windows = attn_windows.reshape(\n- [-1, self.window_size, self.window_size, C])\n+ [x_windows.shape[0], self.window_size, self.window_size, C])\nshifted_x = window_reverse(attn_windows, self.window_size, Hp,\nWp) # B H' W' C\n@@ -314,7 +314,7 @@ class SwinTransformerBlock(nn.Layer):\nif pad_r > 0 or pad_b > 0:\nx = x[:, :H, :W, :]\n- x = x.reshape([B, H * W, C])\n+ x = x.reshape([-1, H * W, C])\n# FFN\nx = shortcut + self.drop_path(x)\n@@ -345,7 +345,7 @@ class PatchMerging(nn.Layer):\nB, L, C = x.shape\nassert L == H * W, \"input feature has wrong size\"\n- x = x.reshape([B, H, W, C])\n+ x = x.reshape([-1, H, W, C])\n# padding\npad_input = (H % 2 == 1) or (W % 2 == 1)\n@@ -357,7 +357,7 @@ class PatchMerging(nn.Layer):\nx2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C\nx3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C\nx = paddle.concat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C\n- x = x.reshape([B, H * W // 4, 4 * C]) # B H/2*W/2 4*C\n+ x = x.reshape([-1, H * W // 4, 4 * C]) # B H/2*W/2 4*C\nx = self.norm(x)\nx = self.reduction(x)\n@@ -664,7 +664,7 @@ class SwinTransformer(nn.Layer):\ndef forward(self, x):\n\"\"\"Forward function.\"\"\"\nx = self.patch_embed(x['image'])\n- _, _, Wh, Ww = x.shape\n+ B, _, Wh, Ww = x.shape\nif self.ape:\n# interpolate the position embedding to the corresponding size\nabsolute_pos_embed = F.interpolate(\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix export_model for swin (#6399)
|
499,339 |
13.07.2022 12:40:43
| -28,800 |
6c59641e92cb5754dcc21e09e1a956fb81e01678
|
[dev] add amp eval
cast dtype in load_pretrain_weight
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -65,6 +65,8 @@ class Trainer(object):\nself.mode = mode.lower()\nself.optimizer = None\nself.is_loaded_weights = False\n+ self.use_amp = self.cfg.get('amp', False)\n+ self.amp_level = self.cfg.get('amp_level', 'O1')\n# build data loader\ncapital_mode = self.mode.capitalize()\n@@ -124,17 +126,6 @@ class Trainer(object):\nelse:\nself.model.load_meanstd(cfg['TestReader']['sample_transforms'])\n- self.use_ema = ('use_ema' in cfg and cfg['use_ema'])\n- if self.use_ema:\n- ema_decay = self.cfg.get('ema_decay', 0.9998)\n- cycle_epoch = self.cfg.get('cycle_epoch', -1)\n- ema_decay_type = self.cfg.get('ema_decay_type', 'threshold')\n- self.ema = ModelEMA(\n- self.model,\n- decay=ema_decay,\n- ema_decay_type=ema_decay_type,\n- cycle_epoch=cycle_epoch)\n-\n# EvalDataset build with BatchSampler to evaluate in single device\n# TODO: multi-device evaluate\nif self.mode == 'eval':\n@@ -162,6 +153,20 @@ class Trainer(object):\nself.pruner = create('UnstructuredPruner')(self.model,\nsteps_per_epoch)\n+ if self.use_amp and self.amp_level == 'O2':\n+ self.model = paddle.amp.decorate(\n+ models=self.model, level=self.amp_level)\n+ self.use_ema = ('use_ema' in cfg and cfg['use_ema'])\n+ if self.use_ema:\n+ ema_decay = self.cfg.get('ema_decay', 0.9998)\n+ cycle_epoch = self.cfg.get('cycle_epoch', -1)\n+ ema_decay_type = self.cfg.get('ema_decay_type', 'threshold')\n+ self.ema = ModelEMA(\n+ self.model,\n+ decay=ema_decay,\n+ ema_decay_type=ema_decay_type,\n+ cycle_epoch=cycle_epoch)\n+\nself._nranks = dist.get_world_size()\nself._local_rank = dist.get_rank()\n@@ -387,13 +392,10 @@ class Trainer(object):\nmodel = paddle.nn.SyncBatchNorm.convert_sync_batchnorm(model)\n# enabel auto mixed precision mode\n- use_amp = self.cfg.get('amp', False)\n- amp_level = self.cfg.get('amp_level', 'O1')\n- if use_amp:\n+ if self.use_amp:\nscaler = paddle.amp.GradScaler(\nenable=self.cfg.use_gpu or self.cfg.use_npu,\ninit_loss_scaling=self.cfg.get('init_loss_scaling', 1024))\n- model = paddle.amp.decorate(models=model, level=amp_level)\n# get distributed model\nif self.cfg.get('fleet', False):\nmodel = fleet.distributed_model(model)\n@@ -438,9 +440,9 @@ class Trainer(object):\nself._compose_callback.on_step_begin(self.status)\ndata['epoch_id'] = epoch_id\n- if use_amp:\n+ if self.use_amp:\nwith paddle.amp.auto_cast(\n- enable=self.cfg.use_gpu, level=amp_level):\n+ enable=self.cfg.use_gpu, level=self.amp_level):\n# model forward\noutputs = model(data)\nloss = outputs['loss']\n@@ -532,6 +534,11 @@ class Trainer(object):\nself.status['step_id'] = step_id\nself._compose_callback.on_step_begin(self.status)\n# forward\n+ if self.use_amp:\n+ with paddle.amp.auto_cast(\n+ enable=self.cfg.use_gpu, level=self.amp_level):\n+ outs = self.model(data)\n+ else:\nouts = self.model(data)\n# update metrics\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/utils.py",
"new_path": "ppdet/modeling/assigners/utils.py",
"diff": "@@ -176,7 +176,8 @@ def compute_max_iou_gt(ious):\ndef generate_anchors_for_grid_cell(feats,\nfpn_strides,\ngrid_cell_size=5.0,\n- grid_cell_offset=0.5):\n+ grid_cell_offset=0.5,\n+ dtype='float32'):\nr\"\"\"\nLike ATSS, generate anchors based on grid size.\nArgs:\n@@ -206,16 +207,15 @@ def generate_anchors_for_grid_cell(feats,\nshift_x - cell_half_size, shift_y - cell_half_size,\nshift_x + cell_half_size, shift_y + cell_half_size\n],\n- axis=-1).astype(feat.dtype)\n- anchor_point = paddle.stack(\n- [shift_x, shift_y], axis=-1).astype(feat.dtype)\n+ axis=-1).astype(dtype)\n+ anchor_point = paddle.stack([shift_x, shift_y], axis=-1).astype(dtype)\nanchors.append(anchor.reshape([-1, 4]))\nanchor_points.append(anchor_point.reshape([-1, 2]))\nnum_anchors_list.append(len(anchors[-1]))\nstride_tensor.append(\npaddle.full(\n- [num_anchors_list[-1], 1], stride, dtype=feat.dtype))\n+ [num_anchors_list[-1], 1], stride, dtype=dtype))\nanchors = paddle.concat(anchors)\nanchors.stop_gradient = True\nanchor_points = paddle.concat(anchor_points)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/ppyoloe_head.py",
"new_path": "ppdet/modeling/heads/ppyoloe_head.py",
"diff": "@@ -160,7 +160,7 @@ class PPYOLOEHead(nn.Layer):\nnum_anchors_list, stride_tensor\n], targets)\n- def _generate_anchors(self, feats=None):\n+ def _generate_anchors(self, feats=None, dtype='float32'):\n# just use in eval time\nanchor_points = []\nstride_tensor = []\n@@ -175,11 +175,9 @@ class PPYOLOEHead(nn.Layer):\nshift_y, shift_x = paddle.meshgrid(shift_y, shift_x)\nanchor_point = paddle.cast(\npaddle.stack(\n- [shift_x, shift_y], axis=-1), dtype='float32')\n+ [shift_x, shift_y], axis=-1), dtype=dtype)\nanchor_points.append(anchor_point.reshape([-1, 2]))\n- stride_tensor.append(\n- paddle.full(\n- [h * w, 1], stride, dtype='float32'))\n+ stride_tensor.append(paddle.full([h * w, 1], stride, dtype=dtype))\nanchor_points = paddle.concat(anchor_points)\nstride_tensor = paddle.concat(stride_tensor)\nreturn anchor_points, stride_tensor\n"
},
{
"change_type": "MODIFY",
"old_path": "tools/eval.py",
"new_path": "tools/eval.py",
"diff": "@@ -77,6 +77,12 @@ def parse_args():\ndefault=False,\nhelp='Whether to save the evaluation results only')\n+ parser.add_argument(\n+ \"--amp\",\n+ action='store_true',\n+ default=False,\n+ help=\"Enable auto mixed precision eval.\")\n+\nargs = parser.parse_args()\nreturn args\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] add amp eval (#6400)
cast dtype in load_pretrain_weight
|
499,339 |
13.07.2022 16:51:15
| -28,800 |
70d00f6fa713383b1b17e249f55575ed57eaa6c0
|
[dev] fix bug in _download_dist
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/utils/download.py",
"new_path": "ppdet/utils/download.py",
"diff": "@@ -393,7 +393,12 @@ def _download(url, path, md5sum=None):\ndef _download_dist(url, path, md5sum=None):\nenv = os.environ\nif 'PADDLE_TRAINERS_NUM' in env and 'PADDLE_TRAINER_ID' in env:\n- trainer_id = int(env['PADDLE_TRAINER_ID'])\n+ # Mainly used to solve the problem of downloading data from\n+ # different machines in the case of multiple machines.\n+ # Different nodes will download data, and the same node\n+ # will only download data once.\n+ # Reference https://github.com/PaddlePaddle/PaddleClas/blob/develop/ppcls/utils/download.py#L108\n+ rank_id_curr_node = int(os.environ.get(\"PADDLE_RANK_IN_NODE\", 0))\nnum_trainers = int(env['PADDLE_TRAINERS_NUM'])\nif num_trainers <= 1:\nreturn _download(url, path, md5sum)\n@@ -406,12 +411,9 @@ def _download_dist(url, path, md5sum=None):\nos.makedirs(path)\nif not osp.exists(fullname):\n- from paddle.distributed import ParallelEnv\n- unique_endpoints = _get_unique_endpoints(ParallelEnv()\n- .trainer_endpoints[:])\nwith open(lock_path, 'w'): # touch\nos.utime(lock_path, None)\n- if ParallelEnv().current_endpoint in unique_endpoints:\n+ if rank_id_curr_node == 0:\n_download(url, path, md5sum)\nos.remove(lock_path)\nelse:\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_train_inference_python.sh",
"new_path": "test_tipc/test_train_inference_python.sh",
"diff": "@@ -262,7 +262,7 @@ else\ncontinue\nfi\n- if [ ${autocast} = \"amp\" ]; then\n+ if [ ${autocast} = \"amp\" ] || [ ${autocast} = \"fp16\" ]; then\nset_autocast=\"--amp\"\nelse\nset_autocast=\" \"\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] fix bug in _download_dist (#6419)
|
499,405 |
13.07.2022 21:44:19
| -28,800 |
5e02a81af77a9a4ecd1e394430c4396b48bc76fd
|
remove one line redundant code
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/layers.py",
"new_path": "ppdet/modeling/layers.py",
"diff": "@@ -749,7 +749,6 @@ class TTFBox(object):\n# batch size is 1\nscores_r = paddle.reshape(scores, [cat, -1])\ntopk_scores, topk_inds = paddle.topk(scores_r, k)\n- topk_scores, topk_inds = paddle.topk(scores_r, k)\ntopk_ys = topk_inds // width\ntopk_xs = topk_inds % width\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
remove one line redundant code (#6424)
|
499,339 |
14.07.2022 14:10:27
| -28,800 |
4a8fe37080d928bf8cc3c9f49946b7d52b2f3974
|
[dev] fix some deadlink in tipc and deploy, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/README_en.md",
"new_path": "deploy/README_en.md",
"diff": "@@ -21,7 +21,7 @@ Use the `tools/export_model.py` script to export the model and the configuration\n# The YOLOv3 model is derived\npython tools/export_model.py -c configs/yolov3/yolov3_mobilenet_v1_roadsign.yml -o weights=output/yolov3_mobilenet_v1_roadsign/best_model.pdparams\n```\n-The prediction model will be exported to the `output_inference/yolov3_mobilenet_v1_roadsign` directory `infer_cfg.yml`, `model.pdiparams`, `model.pdiparams.info`, `model.pdmodel`. For details on model export, please refer to the documentation [Tutorial on Paddle Detection MODEL EXPORT](EXPORT_MODEL_sh.md).\n+The prediction model will be exported to the `output_inference/yolov3_mobilenet_v1_roadsign` directory `infer_cfg.yml`, `model.pdiparams`, `model.pdiparams.info`, `model.pdmodel`. For details on model export, please refer to the documentation [Tutorial on Paddle Detection MODEL EXPORT](./EXPORT_MODEL_en.md).\n### 1.2 Use Paddle Inference to Make Predictions\n* Python deployment supports `CPU`, `GPU` and `XPU` environments, Windows, Linux, and NV Jetson embedded devices. Reference Documentation [Python Deployment](python/README.md)\n@@ -39,7 +39,7 @@ python tools/export_model.py -c configs/yolov3/yolov3_mobilenet_v1_roadsign.yml\n```\nThe prediction model will be exported to the `output_inference/yolov3_darknet53_270e_coco` directory `infer_cfg.yml`, `model.pdiparams`, `model.pdiparams.info`, `model.pdmodel`, `serving_client/` and `serving_server/` folder.\n-For details on model export, please refer to the documentation [Tutorial on Paddle Detection MODEL EXPORT](EXPORT_MODEL_en.md).\n+For details on model export, please refer to the documentation [Tutorial on Paddle Detection MODEL EXPORT](./EXPORT_MODEL_en.md).\n### 2.2 Predictions are made using Paddle Serving\n* [Install PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README.md#installation)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] fix some deadlink in tipc and deploy, test=document_fix (#6431)
|
499,319 |
14.07.2022 14:52:26
| -28,800 |
99f891bebc3113bf6720fc7dbcac2a0e2144b774
|
[doc] deadlinks fix
|
[
{
"change_type": "MODIFY",
"old_path": "configs/mot/README_en.md",
"new_path": "configs/mot/README_en.md",
"diff": "@@ -79,7 +79,7 @@ PaddleDetection implement [JDE](https://github.com/Zhongdao/Towards-Realtime-MOT\n**Notes:**\n- Multi-Object Tracking(MOT) datasets are always used for single category tracking. DeepSORT, JDE and FairMOT are single category MOT models. 'MIX' dataset and it's sub datasets are also single category pedestrian tracking datasets. It can be considered that there are additional IDs ground truth for detection datasets.\n-- In order to train the feature models of more scenes, more datasets are also processed into the same format as the MIX dataset. PaddleDetection Team also provides feature datasets and models of [vehicle tracking](vehicle/readme.md), [head tracking](headtracking21/readme.md) and more general [pedestrian tracking](pedestrian/readme.md). User defined datasets can also be prepared by referring to data preparation [doc](../../docs/tutorials/PrepareMOTDataSet.md).\n+- In order to train the feature models of more scenes, more datasets are also processed into the same format as the MIX dataset. PaddleDetection Team also provides feature datasets and models of [vehicle tracking](vehicle/README.md), [head tracking](headtracking21/README.md) and more general [pedestrian tracking](pedestrian/README.md). User defined datasets can also be prepared by referring to data preparation [doc](../../docs/tutorials/data/PrepareMOTDataSet.md).\n- The multipe category MOT model is [MCFairMOT] (mcfairmot/readme_cn.md), and the multi category dataset is the integrated version of VisDrone dataset. Please refer to the doc of [MCFairMOT](mcfairmot/README.md).\n- The Multi-Target Multi-Camera Tracking (MTMCT) model is [AIC21 MTMCT](https://www.aicitychallenge.org)(CityFlow) Multi-Camera Vehicle Tracking dataset. The dataset and model can refer to the doc of [MTMCT](mtmct/README.md)\n"
},
{
"change_type": "MODIFY",
"old_path": "docs/tutorials/GETTING_STARTED.md",
"new_path": "docs/tutorials/GETTING_STARTED.md",
"diff": "@@ -11,7 +11,7 @@ instructions](INSTALL_cn.md).\n## Data preparation\n-- Please refer to [PrepareDetDataSet](PrepareDetDataSet_en.md) for data preparation\n+- Please refer to [PrepareDetDataSet](./data/PrepareDetDataSet_en.md) for data preparation\n- Please set the data path for data configuration file in ```configs/datasets```\n## Training & Evaluation & Inference\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[doc] deadlinks fix (#6434)
|
499,339 |
14.07.2022 19:38:35
| -28,800 |
a2b3d3c0a1ca427db8296d6ece40e116b085a696
|
[TIPC] add dist train txt, test=document_fix
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/keypoint/tinypose_128x96_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:tinypose_128x96\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=420\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=512\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/keypoint/tinypose_128x96.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c test_tipc/configs/keypoint/tinypose_128x96.yml -o\n+pact_train:tools/train.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c test_tipc/configs/keypoint/tinypose_128x96.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/keypoint/tinypose_128x96.pdparams\n+norm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml -o\n+pact_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c test_tipc/configs/keypoint/tinypose_128x96.yml --slim_config configs/slim/post_quant/tinypose_128x96_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/keypoint_infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1|2\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:mask_rcnn_r50_fpn_1x_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=12\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=1\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_1x_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml -o\n+pact_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_1x_coco.pdparams\n+norm_export:tools/export_model.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml -o\n+pact_export:tools/export_model.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export_onnx:null\n+kl_quant_export:tools/post_quant.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+--trt_max_shape:1600\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/picodet/picodet_lcnet_1_5x_416_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:picodet_lcnet_1_5x_416_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=300\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=80\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/picodet_lcnet_1_5x_416_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml -o\n+pact_train:tools/train.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/picodet_lcnet_1_5x_416_coco.pdparams\n+norm_export:tools/export_model.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml -o\n+pact_export:tools/export_model.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/picodet/picodet_s_320_coco_lcnet_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:picodet_s_320_coco_lcnet\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=300\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=128\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/picodet_s_320_coco_lcnet.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/picodet/picodet_s_320_coco_lcnet.yml -o\n+pact_train:tools/train.py -c configs/picodet/picodet_s_320_coco_lcnet.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c configs/picodet/picodet_s_320_coco_lcnet.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/picodet/picodet_s_320_coco_lcnet.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/picodet_s_320_coco_lcnet.pdparams\n+norm_export:tools/export_model.py -c configs/picodet/picodet_s_320_coco_lcnet.yml -o\n+pact_export:tools/export_model.py -c configs/picodet/picodet_s_320_coco_lcnet.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c configs/picodet/picodet_s_320_coco_lcnet.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/picodet/picodet_s_320_coco_lcnet.yml --slim_config _template_kl_quant -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/picodet/picodet_s_320_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:picodet_s_320_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=300\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=128\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/picodet_s_320_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml -o\n+pact_train:tools/train.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/picodet_s_320_coco.pdparams\n+norm_export:tools/export_model.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml -o\n+pact_export:tools/export_model.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/ppyolo/ppyolo_mbv3_large_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:ppyolo_mbv3_large_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=405\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=24\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml -o\n+pact_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/quant/ppyolo_mbv3_large_qat.yml -o\n+fpgm_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/prune/ppyolo_mbv3_large_prune_fpgm.yml -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams\n+norm_export:tools/export_model.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml -o\n+pact_export:tools/export_model.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/quant/ppyolo_mbv3_large_qat.yml -o\n+fpgm_export:tools/export_model.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/prune/ppyolo_mbv3_large_prune_fpgm.yml -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml --slim_config configs/slim/post_quant/ppyolo_mbv3_large_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1|2\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n+===========================infer_benchmark_params===========================\n+numpy_infer_input:3x320x320.npy\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:ppyolo_r50vd_dcn_1x_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=405\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=24\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml -o\n+pact_train:tools/train.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml --slim_config configs/slim/quant/ppyolo_r50vd_qat_pact.yml -o\n+fpgm_train:tools/train.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml --slim_config configs/slim/prune/ppyolo_r50vd_prune_fpgm.yml -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams\n+norm_export:tools/export_model.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml -o\n+pact_export:tools/export_model.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml --slim_config configs/slim/quant/ppyolo_r50vd_qat_pact.yml -o\n+fpgm_export:tools/export_model.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml --slim_config configs/slim/prune/ppyolo_r50vd_prune_fpgm.yml -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml --slim_config configs/slim/post_quant/ppyolo_r50vd_dcn_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1|2\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n+===========================infer_benchmark_params===========================\n+numpy_infer_input:3x608x608.npy\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/ppyolo/ppyolo_tiny_650e_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:ppyolo_tiny_650e_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=650\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=32\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolo_tiny_650e_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml -o\n+pact_train:tools/train.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/ppyolo_tiny_650e_coco.pdparams\n+norm_export:tools/export_model.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml -o\n+pact_export:tools/export_model.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1|2\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n+===========================infer_benchmark_params===========================\n+numpy_infer_input:3x320x320.npy\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:ppyolov2_r50vd_dcn_365e_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=365\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=12\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolov2_r50vd_dcn_365e_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml -o\n+pact_train:tools/train.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --slim_config _template_pact -o\n+fpgm_train:tools/train.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --slim_config _template_fpgm -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/ppyolov2_r50vd_dcn_365e_coco.pdparams\n+norm_export:tools/export_model.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml -o\n+pact_export:tools/export_model.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --slim_config _template_pact -o\n+fpgm_export:tools/export_model.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --slim_config _template_fpgm -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1|2\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n+===========================infer_benchmark_params===========================\n+numpy_infer_input:3x640x640.npy\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/yolov3/yolov3_darknet53_270e_coco_train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt",
"diff": "+===========================train_params===========================\n+model_name:yolov3_darknet53_270e_coco\n+python:python3.7\n+gpu_list:192.168.0.1,192.168.0.2;0,1\n+use_gpu:True\n+auto_cast:null\n+epoch:lite_train_lite_infer=1|lite_train_whole_infer=1|whole_train_whole_infer=270\n+save_dir:null\n+TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_train_whole_infer=8\n+pretrain_weights:https://paddledet.bj.bcebos.com/models/yolov3_darknet53_270e_coco.pdparams\n+trained_model_name:model_final.pdparams\n+train_infer_img_dir:./dataset/coco/test2017/\n+filename:null\n+##\n+trainer:norm_train\n+norm_train:tools/train.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml -o\n+pact_train:tools/train.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml --slim_config configs/slim/quant/yolov3_darknet_qat.yml -o\n+fpgm_train:tools/train.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml --slim_config configs/slim/prune/yolov3_darknet_prune_fpgm.yml -o\n+distill_train:null\n+null:null\n+null:null\n+##\n+===========================eval_params===========================\n+eval:tools/eval.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml -o\n+null:null\n+##\n+===========================infer_params===========================\n+--output_dir:./output_inference\n+weights:https://paddledet.bj.bcebos.com/models/yolov3_darknet53_270e_coco.pdparams\n+norm_export:tools/export_model.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml -o\n+pact_export:tools/export_model.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml --slim_config configs/slim/quant/yolov3_darknet_qat.yml -o\n+fpgm_export:tools/export_model.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml --slim_config configs/slim/prune/yolov3_darknet_prune_fpgm.yml -o\n+distill_export:null\n+export1:null\n+export2:null\n+kl_quant_export:tools/post_quant.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml --slim_config configs/slim/post_quant/yolov3_darknet53_ptq.yml -o\n+##\n+infer_mode:norm\n+infer_quant:False\n+inference:./deploy/python/infer.py\n+--device:cpu\n+--enable_mkldnn:False\n+--cpu_threads:4\n+--batch_size:1|2\n+--use_tensorrt:null\n+--run_mode:paddle\n+--model_dir:\n+--image_dir:./dataset/coco/test2017/\n+--save_log_path:null\n+--run_benchmark:False\n+null:null\n\\ No newline at end of file\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] add dist train txt, test=document_fix (#6440)
|
499,348 |
14.07.2022 21:56:56
| -28,800 |
63dc4c4afe129c59c80ac94122ff3f0ba75404e6
|
update pphumandocs&annodocs; test=document_fix
|
[
{
"change_type": "ADD",
"old_path": "docs/images/pphumanv2.png",
"new_path": "docs/images/pphumanv2.png",
"diff": "Binary files /dev/null and b/docs/images/pphumanv2.png differ\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update pphumandocs&annodocs; test=document_fix (#6442)
|
499,319 |
15.07.2022 20:27:17
| -28,800 |
0910e9882985d8100dcdc6b8055dec0fe554f496
|
add use_checkpoint and use_alpha for cspresnet
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "configs/visdrone/ppyoloe_crn_s_80e_visdrone_use_checkpoint.yml",
"diff": "+_BASE_: [\n+ '../datasets/visdrone_detection.yml',\n+ '../runtime.yml',\n+ '../ppyoloe/_base_/optimizer_300e.yml',\n+ '../ppyoloe/_base_/ppyoloe_crn.yml',\n+ '../ppyoloe/_base_/ppyoloe_reader.yml',\n+]\n+log_iter: 100\n+snapshot_epoch: 10\n+weights: output/ppyoloe_crn_s_80e_visdrone_use_checkpoint/model_final\n+\n+pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_300e_coco.pdparams\n+depth_mult: 0.33\n+width_mult: 0.50\n+\n+TrainReader:\n+ batch_size: 8\n+\n+epoch: 80\n+LearningRate:\n+ base_lr: 0.01\n+ schedulers:\n+ - !CosineDecay\n+ max_epochs: 96\n+ - !LinearWarmup\n+ start_factor: 0.\n+ epochs: 1\n+\n+CSPResNet:\n+ use_checkpoint: True\n+ use_alpha: True\n+\n+# when use_checkpoint\n+use_fused_allreduce_gradients: True\n+\n+PPYOLOEHead:\n+ static_assigner_epoch: -1\n+ nms:\n+ name: MultiClassNMS\n+ nms_top_k: 10000\n+ keep_top_k: 500\n+ score_threshold: 0.01\n+ nms_threshold: 0.6\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -49,6 +49,8 @@ from ppdet.utils import profiler\nfrom .callbacks import Callback, ComposeCallback, LogPrinter, Checkpointer, WiferFaceEval, VisualDLWriter, SniperProposalsGenerator, WandbCallback\nfrom .export_utils import _dump_infer_config, _prune_input_spec\n+from paddle.distributed.fleet.utils.hybrid_parallel_util import fused_allreduce_gradients\n+\nfrom ppdet.utils.logger import setup_logger\nlogger = setup_logger('ppdet.engine')\n@@ -152,7 +154,6 @@ class Trainer(object):\nif self.cfg.get('unstructured_prune'):\nself.pruner = create('UnstructuredPruner')(self.model,\nsteps_per_epoch)\n-\nif self.use_amp and self.amp_level == 'O2':\nself.model = paddle.amp.decorate(\nmodels=self.model, level=self.amp_level)\n@@ -426,6 +427,9 @@ class Trainer(object):\nself._compose_callback.on_train_begin(self.status)\n+ use_fused_allreduce_gradients = self.cfg[\n+ 'use_fused_allreduce_gradients'] if 'use_fused_allreduce_gradients' in self.cfg else False\n+\nfor epoch_id in range(self.start_epoch, self.cfg.epoch):\nself.status['mode'] = 'train'\nself.status['epoch_id'] = epoch_id\n@@ -441,7 +445,23 @@ class Trainer(object):\ndata['epoch_id'] = epoch_id\nif self.use_amp:\n- with paddle.amp.auto_cast(\n+ if isinstance(\n+ model, paddle.\n+ DataParallel) and use_fused_allreduce_gradients:\n+ with model.no_sync():\n+ with amp.auto_cast(\n+ enable=self.cfg.use_gpus,\n+ level=self.amp_level):\n+ # model forward\n+ outputs = model(data)\n+ loss = outputs['loss']\n+ # model backward\n+ scaled_loss = scaler.scale(loss)\n+ scaled_loss.backward()\n+ fused_allreduce_gradients(\n+ list(model.parameters()), None)\n+ else:\n+ with amp.auto_cast(\nenable=self.cfg.use_gpu, level=self.amp_level):\n# model forward\noutputs = model(data)\n@@ -451,6 +471,19 @@ class Trainer(object):\nscaled_loss.backward()\n# in dygraph mode, optimizer.minimize is equal to optimizer.step\nscaler.minimize(self.optimizer, scaled_loss)\n+\n+ else:\n+ if isinstance(\n+ model, paddle.\n+ DataParallel) and use_fused_allreduce_gradients:\n+ with model.no_sync():\n+ # model forward\n+ outputs = model(data)\n+ loss = outputs['loss']\n+ # model backward\n+ loss.backward()\n+ fused_allreduce_gradients(\n+ list(model.parameters()), None)\nelse:\n# model forward\noutputs = model(data)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/cspresnet.py",
"new_path": "ppdet/modeling/backbones/cspresnet.py",
"diff": "@@ -21,6 +21,7 @@ import paddle.nn as nn\nimport paddle.nn.functional as F\nfrom paddle import ParamAttr\nfrom paddle.regularizer import L2Decay\n+from paddle.nn.initializer import Constant\nfrom ppdet.modeling.ops import get_act_fn\nfrom ppdet.core.workspace import register, serializable\n@@ -65,7 +66,7 @@ class ConvBNLayer(nn.Layer):\nclass RepVggBlock(nn.Layer):\n- def __init__(self, ch_in, ch_out, act='relu'):\n+ def __init__(self, ch_in, ch_out, act='relu', alpha=False):\nsuper(RepVggBlock, self).__init__()\nself.ch_in = ch_in\nself.ch_out = ch_out\n@@ -75,10 +76,20 @@ class RepVggBlock(nn.Layer):\nch_in, ch_out, 1, stride=1, padding=0, act=None)\nself.act = get_act_fn(act) if act is None or isinstance(act, (\nstr, dict)) else act\n+ if alpha:\n+ self.alpha = self.create_parameter(\n+ shape=[1],\n+ attr=ParamAttr(initializer=Constant(value=1.)),\n+ dtype=\"float32\")\n+ else:\n+ self.alpha = None\ndef forward(self, x):\nif hasattr(self, 'conv'):\ny = self.conv(x)\n+ else:\n+ if self.alpha:\n+ y = self.conv1(x) + self.alpha * self.conv2(x)\nelse:\ny = self.conv1(x) + self.conv2(x)\ny = self.act(y)\n@@ -102,6 +113,10 @@ class RepVggBlock(nn.Layer):\ndef get_equivalent_kernel_bias(self):\nkernel3x3, bias3x3 = self._fuse_bn_tensor(self.conv1)\nkernel1x1, bias1x1 = self._fuse_bn_tensor(self.conv2)\n+ if self.alpha:\n+ return kernel3x3 + self.alpha * self._pad_1x1_to_3x3_tensor(\n+ kernel1x1), bias3x3 + self.alpha * bias1x1\n+ else:\nreturn kernel3x3 + self._pad_1x1_to_3x3_tensor(\nkernel1x1), bias3x3 + bias1x1\n@@ -126,11 +141,16 @@ class RepVggBlock(nn.Layer):\nclass BasicBlock(nn.Layer):\n- def __init__(self, ch_in, ch_out, act='relu', shortcut=True):\n+ def __init__(self,\n+ ch_in,\n+ ch_out,\n+ act='relu',\n+ shortcut=True,\n+ use_alpha=False):\nsuper(BasicBlock, self).__init__()\nassert ch_in == ch_out\nself.conv1 = ConvBNLayer(ch_in, ch_out, 3, stride=1, padding=1, act=act)\n- self.conv2 = RepVggBlock(ch_out, ch_out, act=act)\n+ self.conv2 = RepVggBlock(ch_out, ch_out, act=act, alpha=use_alpha)\nself.shortcut = shortcut\ndef forward(self, x):\n@@ -167,7 +187,8 @@ class CSPResStage(nn.Layer):\nn,\nstride,\nact='relu',\n- attn='eca'):\n+ attn='eca',\n+ use_alpha=False):\nsuper(CSPResStage, self).__init__()\nch_mid = (ch_in + ch_out) // 2\n@@ -180,8 +201,11 @@ class CSPResStage(nn.Layer):\nself.conv2 = ConvBNLayer(ch_mid, ch_mid // 2, 1, act=act)\nself.blocks = nn.Sequential(*[\nblock_fn(\n- ch_mid // 2, ch_mid // 2, act=act, shortcut=True)\n- for i in range(n)\n+ ch_mid // 2,\n+ ch_mid // 2,\n+ act=act,\n+ shortcut=True,\n+ use_alpha=use_alpha) for i in range(n)\n])\nif attn:\nself.attn = EffectiveSELayer(ch_mid, act='hardsigmoid')\n@@ -216,8 +240,12 @@ class CSPResNet(nn.Layer):\nuse_large_stem=False,\nwidth_mult=1.0,\ndepth_mult=1.0,\n- trt=False):\n+ trt=False,\n+ use_checkpoint=False,\n+ use_alpha=False,\n+ **args):\nsuper(CSPResNet, self).__init__()\n+ self.use_checkpoint = use_checkpoint\nchannels = [max(round(c * width_mult), 1) for c in channels]\nlayers = [max(round(l * depth_mult), 1) for l in layers]\nact = get_act_fn(\n@@ -255,18 +283,29 @@ class CSPResNet(nn.Layer):\nn = len(channels) - 1\nself.stages = nn.Sequential(*[(str(i), CSPResStage(\n- BasicBlock, channels[i], channels[i + 1], layers[i], 2, act=act))\n- for i in range(n)])\n+ BasicBlock,\n+ channels[i],\n+ channels[i + 1],\n+ layers[i],\n+ 2,\n+ act=act,\n+ use_alpha=use_alpha)) for i in range(n)])\nself._out_channels = channels[1:]\n- self._out_strides = [4, 8, 16, 32]\n+ self._out_strides = [4 * 2**i for i in range(n)]\nself.return_idx = return_idx\n+ if use_checkpoint:\n+ paddle.seed(0)\ndef forward(self, inputs):\nx = inputs['image']\nx = self.stem(x)\nouts = []\nfor idx, stage in enumerate(self.stages):\n+ if self.use_checkpoint and self.training:\n+ x = paddle.distributed.fleet.utils.recompute(\n+ stage, x, **{\"preserve_rng_state\": True})\n+ else:\nx = stage(x)\nif idx in self.return_idx:\nouts.append(x)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add use_checkpoint and use_alpha for cspresnet (#6428)
|
499,298 |
18.07.2022 10:36:26
| -28,800 |
b9a2d36d656285ce510fab368138257800cb66a2
|
fix amp training
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -449,7 +449,7 @@ class Trainer(object):\nmodel, paddle.\nDataParallel) and use_fused_allreduce_gradients:\nwith model.no_sync():\n- with amp.auto_cast(\n+ with paddle.amp.auto_cast(\nenable=self.cfg.use_gpus,\nlevel=self.amp_level):\n# model forward\n@@ -461,7 +461,7 @@ class Trainer(object):\nfused_allreduce_gradients(\nlist(model.parameters()), None)\nelse:\n- with amp.auto_cast(\n+ with paddle.amp.auto_cast(\nenable=self.cfg.use_gpu, level=self.amp_level):\n# model forward\noutputs = model(data)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix amp training (#6456)
|
499,298 |
18.07.2022 17:08:08
| -28,800 |
e6ad10e5cf86676c898fe27af19dc991fc4d98f4
|
fix ids2names in plot_tracking_dict
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipeline.py",
"new_path": "deploy/pipeline/pipeline.py",
"diff": "@@ -966,6 +966,7 @@ class PipePredictor(object):\nonline_scores,\nframe_id=frame_id,\nfps=fps,\n+ ids2names=self.mot_predictor.pred_config.labels,\ndo_entrance_counting=self.do_entrance_counting,\ndo_break_in_counting=self.do_break_in_counting,\nentrance=entrance,\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot/visualize.py",
"new_path": "deploy/pptracking/python/mot/visualize.py",
"diff": "@@ -191,7 +191,7 @@ def plot_tracking_dict(image,\nscores_dict,\nframe_id=0,\nfps=0.,\n- ids2names=['pedestrian'],\n+ ids2names=[],\ndo_entrance_counting=False,\ndo_break_in_counting=False,\nentrance=None,\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot_sde_infer.py",
"new_path": "deploy/pptracking/python/mot_sde_infer.py",
"diff": "@@ -512,14 +512,15 @@ class SDE_Detector(Detector):\nonline_ids,\nonline_scores,\nframe_id=frame_id,\n- ids2names=[])\n+ ids2names=ids2names)\nelse:\nim = plot_tracking(\nframe,\nonline_tlwhs,\nonline_ids,\nonline_scores,\n- frame_id=frame_id)\n+ frame_id=frame_id,\n+ ids2names=ids2names)\nsave_dir = os.path.join(self.output_dir, seq_name)\nif not os.path.exists(save_dir):\nos.makedirs(save_dir)\n@@ -632,6 +633,7 @@ class SDE_Detector(Detector):\nonline_scores,\nframe_id=frame_id,\nfps=fps,\n+ ids2names=ids2names,\ndo_entrance_counting=self.do_entrance_counting,\nentrance=entrance)\nelse:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/mot_sde_infer.py",
"new_path": "deploy/python/mot_sde_infer.py",
"diff": "@@ -359,14 +359,15 @@ class SDE_Detector(Detector):\nonline_ids,\nonline_scores,\nframe_id=frame_id,\n- ids2names=[])\n+ ids2names=ids2names)\nelse:\nim = plot_tracking(\nframe,\nonline_tlwhs,\nonline_ids,\nonline_scores,\n- frame_id=frame_id)\n+ frame_id=frame_id,\n+ ids2names=ids2names)\nsave_dir = os.path.join(self.output_dir, seq_name)\nif not os.path.exists(save_dir):\nos.makedirs(save_dir)\n@@ -431,7 +432,8 @@ class SDE_Detector(Detector):\nonline_ids,\nonline_scores,\nframe_id=frame_id,\n- fps=fps)\n+ fps=fps,\n+ ids2names=ids2names)\nelse:\n# use ByteTracker, support multiple class\nfor cls_id in range(num_classes):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix ids2names in plot_tracking_dict (#6466)
|
499,339 |
25.07.2022 11:11:34
| -28,800 |
3645208e752ae8cce1bac934d3453915bb99434b
|
[dev] alter ppyoloe nms params
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"diff": "@@ -39,7 +39,7 @@ PPYOLOEHead:\nbeta: 6.0\nnms:\nname: MultiClassNMS\n- nms_top_k: 10000\n+ nms_top_k: 1000\nkeep_top_k: 300\nscore_threshold: 0.01\nnms_threshold: 0.7\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/ppyoloe_crn_l_36e_coco_xpu.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_l_36e_coco_xpu.yml",
"diff": "@@ -64,6 +64,6 @@ PPYOLOEHead:\nnms:\nname: MultiClassNMS\nnms_top_k: 1000\n- keep_top_k: 100\n+ keep_top_k: 300\nscore_threshold: 0.01\n- nms_threshold: 0.6\n+ nms_threshold: 0.7\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/ppyoloe_crn_s_400e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_s_400e_coco.yml",
"diff": "@@ -17,13 +17,9 @@ width_mult: 0.50\nTrainReader:\nbatch_size: 32\n-LearningRate:\n- base_lr: 0.04\n-\n-\nepoch: 400\nLearningRate:\n- base_lr: 0.025\n+ base_lr: 0.04\nschedulers:\n- !CosineDecay\nmax_epochs: 480\n@@ -44,7 +40,7 @@ PPYOLOEHead:\nstatic_assigner_epoch: 133\nnms:\nname: MultiClassNMS\n- nms_top_k: 10000\n+ nms_top_k: 1000\nkeep_top_k: 300\nscore_threshold: 0.01\nnms_threshold: 0.7\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] alter ppyoloe nms params (#6491)
|
499,298 |
25.07.2022 17:03:55
| -28,800 |
324b0b9961d19679192431ade74f6ce956e67f7f
|
fix cpp infer of jdetracker
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/src/tracker.cc",
"new_path": "deploy/cpp/src/tracker.cc",
"diff": "@@ -58,8 +58,8 @@ bool JDETracker::update(const cv::Mat &dets, const cv::Mat &emb, std::vector<Tra\nTrajectoryPool candidates(dets.rows);\nfor (int i = 0; i < dets.rows; ++i)\n{\n- float score = *dets.ptr<float>(i, 4);\n- const cv::Mat <rb_ = dets(cv::Rect(0, i, 4, 1));\n+ float score = *dets.ptr<float>(i, 1);\n+ const cv::Mat <rb_ = dets(cv::Rect(2, i, 4, 1));\ncv::Vec4f ltrb = mat2vec4f(ltrb_);\nconst cv::Mat &embedding = emb(cv::Rect(0, i, emb.cols, 1));\ncandidates[i] = Trajectory(ltrb, score, embedding);\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/cpp/src/tracker.cc",
"new_path": "deploy/pptracking/cpp/src/tracker.cc",
"diff": "@@ -56,8 +56,8 @@ bool JDETracker::update(const cv::Mat &dets,\n++timestamp;\nTrajectoryPool candidates(dets.rows);\nfor (int i = 0; i < dets.rows; ++i) {\n- float score = *dets.ptr<float>(i, 4);\n- const cv::Mat <rb_ = dets(cv::Rect(0, i, 4, 1));\n+ float score = *dets.ptr<float>(i, 1);\n+ const cv::Mat <rb_ = dets(cv::Rect(2, i, 4, 1));\ncv::Vec4f ltrb = mat2vec4f(ltrb_);\nconst cv::Mat &embedding = emb(cv::Rect(0, i, emb.cols, 1));\ncandidates[i] = Trajectory(ltrb, score, embedding);\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix cpp infer of jdetracker (#6500)
|
499,299 |
25.07.2022 23:20:52
| -28,800 |
190e237b2114aef65a26045f78faadb6a6744471
|
fix box filter when box_num > 0 but with no target class
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/det_keypoint_unite_infer.py",
"new_path": "deploy/python/det_keypoint_unite_infer.py",
"diff": "@@ -37,8 +37,15 @@ KEYPOINT_SUPPORT_MODELS = {\ndef predict_with_given_det(image, det_res, keypoint_detector,\nkeypoint_batch_size, run_benchmark):\n+ keypoint_res = {}\n+\nrec_images, records, det_rects = keypoint_detector.get_person_from_rect(\nimage, det_res)\n+\n+ if len(det_rects) == 0:\n+ keypoint_res['keypoint'] = [[], []]\n+ return keypoint_res\n+\nkeypoint_vector = []\nscore_vector = []\n@@ -47,7 +54,6 @@ def predict_with_given_det(image, det_res, keypoint_detector,\nrec_images, run_benchmark, repeats=10, visual=False)\nkeypoint_vector, score_vector = translate_to_ori_images(keypoint_results,\nnp.array(records))\n- keypoint_res = {}\nkeypoint_res['keypoint'] = [\nkeypoint_vector.tolist(), score_vector.tolist()\n] if len(keypoint_vector) > 0 else [[], []]\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix box filter when box_num > 0 but with no target class (#6506)
|
499,333 |
01.08.2022 10:25:44
| -28,800 |
06c8cf7e5a75be43c51323a6c21e21af291e5728
|
fix voc save_result in infer
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/source/dataset.py",
"new_path": "ppdet/data/source/dataset.py",
"diff": "@@ -208,6 +208,10 @@ class ImageFolder(DetDataset):\nself.image_dir = images\nself.roidbs = self._load_images()\n+ def get_label_list(self):\n+ # Only VOC dataset needs label list in ImageFold\n+ return self.anno_path\n+\n@register\nclass CommonDataset(object):\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -287,12 +287,18 @@ class Trainer(object):\nsave_prediction_only=save_prediction_only)\n]\nelif self.cfg.metric == 'VOC':\n+ output_eval = self.cfg['output_eval'] \\\n+ if 'output_eval' in self.cfg else None\n+ save_prediction_only = self.cfg.get('save_prediction_only', False)\n+\nself._metrics = [\nVOCMetric(\nlabel_list=self.dataset.get_label_list(),\nclass_num=self.cfg.num_classes,\nmap_type=self.cfg.map_type,\n- classwise=classwise)\n+ classwise=classwise,\n+ output_eval=output_eval,\n+ save_prediction_only=save_prediction_only)\n]\nelif self.cfg.metric == 'WiderFace':\nmulti_scale = self.cfg.multi_scale_eval if 'multi_scale_eval' in self.cfg else True\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/metrics/metrics.py",
"new_path": "ppdet/metrics/metrics.py",
"diff": "@@ -225,7 +225,9 @@ class VOCMetric(Metric):\nmap_type='11point',\nis_bbox_normalized=False,\nevaluate_difficult=False,\n- classwise=False):\n+ classwise=False,\n+ output_eval=None,\n+ save_prediction_only=False):\nassert os.path.isfile(label_list), \\\n\"label_list {} not a file\".format(label_list)\nself.clsid2catid, self.catid2name = get_categories('VOC', label_list)\n@@ -233,6 +235,8 @@ class VOCMetric(Metric):\nself.overlap_thresh = overlap_thresh\nself.map_type = map_type\nself.evaluate_difficult = evaluate_difficult\n+ self.output_eval = output_eval\n+ self.save_prediction_only = save_prediction_only\nself.detection_map = DetectionMAP(\nclass_num=class_num,\noverlap_thresh=overlap_thresh,\n@@ -245,6 +249,7 @@ class VOCMetric(Metric):\nself.reset()\ndef reset(self):\n+ self.results = {'bbox': [], 'score': [], 'label': []}\nself.detection_map.reset()\ndef update(self, inputs, outputs):\n@@ -256,8 +261,15 @@ class VOCMetric(Metric):\nbbox_lengths = outputs['bbox_num'].numpy() if isinstance(\noutputs['bbox_num'], paddle.Tensor) else outputs['bbox_num']\n+ self.results['bbox'].append(bboxes.tolist())\n+ self.results['score'].append(scores.tolist())\n+ self.results['label'].append(labels.tolist())\n+\nif bboxes.shape == (1, 1) or bboxes is None:\nreturn\n+ if self.save_prediction_only:\n+ return\n+\ngt_boxes = inputs['gt_bbox']\ngt_labels = inputs['gt_class']\ndifficults = inputs['difficult'] if not self.evaluate_difficult \\\n@@ -294,6 +306,15 @@ class VOCMetric(Metric):\nbbox_idx += bbox_num\ndef accumulate(self):\n+ output = \"bbox.json\"\n+ if self.output_eval:\n+ output = os.path.join(self.output_eval, output)\n+ with open(output, 'w') as f:\n+ json.dump(self.results, f)\n+ logger.info('The bbox result is saved to bbox.json.')\n+ if self.save_prediction_only:\n+ return\n+\nlogger.info(\"Accumulating evaluatation results...\")\nself.detection_map.accumulate()\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix voc save_result in infer (#6547)
|
499,333 |
02.08.2022 17:46:12
| -28,800 |
34166cd41551d6b9395e63075321ded209bbf5ad
|
update example, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "README_en.md",
"new_path": "README_en.md",
"diff": "@@ -432,19 +432,15 @@ The comparison between COCO mAP and FPS on Qualcomm Snapdragon 865 processor of\n### [Industrial tutorial examples](./industrial_tutorial/README.md)\n+- [Intelligent fitness recognition based on PP-TinyPose Plus](https://aistudio.baidu.com/aistudio/projectdetail/4385813)\n+\n- [Road litter detection based on PP-PicoDet Plus](https://aistudio.baidu.com/aistudio/projectdetail/3561097)\n- [Communication tower detection based on PP-PicoDet and deployment on Android](https://aistudio.baidu.com/aistudio/projectdetail/3561097)\n-- [Tile surface defect detection based on Faster-RCNN](https://aistudio.baidu.com/aistudio/projectdetail/2571419)\n-\n-- [PCB defect detection based on PaddleDetection](https://aistudio.baidu.com/aistudio/projectdetail/2367089)\n-\n- [Visitor flow statistics based on FairMOT](https://aistudio.baidu.com/aistudio/projectdetail/2421822)\n-- [Falling detection based on YOLOv3](https://aistudio.baidu.com/aistudio/projectdetail/2500639)\n-\n-- [Compliance detection based on human key point detection](https://aistudio.baidu.com/aistudio/projectdetail/4061642?contributionType=1)\n+- [More examples](./industrial_tutorial/README.md)\n## <img title=\"\" src=\"https://user-images.githubusercontent.com/48054808/157836473-1cf451fa-f01f-4148-ba68-b6d06d5da2f9.png\" alt=\"\" width=\"20\"> Applications\n"
},
{
"change_type": "MODIFY",
"old_path": "requirements.txt",
"new_path": "requirements.txt",
"diff": "tqdm\n-typeguard ; python_version >= '3.4'\n-visualdl>=2.1.0 ; python_version <= '3.7'\n+typeguard\n+visualdl>=2.2.0\nopencv-python\nPyYAML\nshapely\n@@ -8,8 +8,7 @@ scipy\nterminaltables\nCython\npycocotools\n-#xtcocotools==1.6 #only for crowdpose\n-setuptools>=42.0.0\n+setuptools\n# for vehicleplate\npyclipper\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update example, test=document_fix (#6567)
|
499,299 |
02.08.2022 17:48:02
| -28,800 |
14987f65ec199d5916702020a8451ecff8ff9a8c
|
fix training error for picodet_s_192_lcnet_pedestrian
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/application/pedestrian_detection/picodet_s_192_lcnet_pedestrian.yml",
"new_path": "configs/picodet/application/pedestrian_detection/picodet_s_192_lcnet_pedestrian.yml",
"diff": "@@ -56,7 +56,7 @@ PicoHeadV2:\nuse_align_head: True\nstatic_assigner:\nname: ATSSAssigner\n- topk: 9\n+ topk: 4\nforce_gt_matching: False\nassigner:\nname: TaskAlignedAssigner\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix training error for picodet_s_192_lcnet_pedestrian (#6562)
|
499,319 |
03.08.2022 14:34:53
| -28,800 |
b276610c3da6d2c46532549b7a08600733db9909
|
PPYOLOE fix out_shape, reset return_idx
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/cspresnet.py",
"new_path": "ppdet/modeling/backbones/cspresnet.py",
"diff": "@@ -235,7 +235,7 @@ class CSPResNet(nn.Layer):\nlayers=[3, 6, 6, 3],\nchannels=[64, 128, 256, 512, 1024],\nact='swish',\n- return_idx=[0, 1, 2, 3, 4],\n+ return_idx=[1, 2, 3],\ndepth_wise=False,\nuse_large_stem=False,\nwidth_mult=1.0,\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
PPYOLOE fix out_shape, reset return_idx (#6561)
|
499,304 |
04.08.2022 14:34:51
| -28,800 |
3e4d5697d9946db21ef10b933658e59c9e97ae1b
|
add PP-YOLOE Auto Compression demo
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/configs/ppyoloe_l_qat_dis.yaml",
"diff": "+\n+Global:\n+ reader_config: configs/yolo_reader.yml\n+ input_list: ['image', 'scale_factor']\n+ arch: YOLO\n+ Evaluation: True\n+ model_dir: ./ppyoloe_crn_l_300e_coco\n+ model_filename: model.pdmodel\n+ params_filename: model.pdiparams\n+\n+Distillation:\n+ alpha: 1.0\n+ loss: soft_label\n+\n+Quantization:\n+ use_pact: true\n+ activation_quantize_type: 'moving_average_abs_max'\n+ quantize_op_types:\n+ - conv2d\n+ - depthwise_conv2d\n+\n+TrainConfig:\n+ train_iter: 5000\n+ eval_iter: 1000\n+ learning_rate:\n+ type: CosineAnnealingDecay\n+ learning_rate: 0.00003\n+ T_max: 6000\n+ optimizer_builder:\n+ optimizer:\n+ type: SGD\n+ weight_decay: 4.0e-05\n+\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/configs/yolo_reader.yml",
"diff": "+metric: COCO\n+num_classes: 80\n+\n+# Datset configuration\n+TrainDataset:\n+ !COCODataSet\n+ image_dir: train2017\n+ anno_path: annotations/instances_train2017.json\n+ dataset_dir: dataset/coco/\n+\n+EvalDataset:\n+ !COCODataSet\n+ image_dir: val2017\n+ anno_path: annotations/instances_val2017.json\n+ dataset_dir: dataset/coco/\n+\n+worker_num: 0\n+\n+# preprocess reader in test\n+EvalReader:\n+ sample_transforms:\n+ - Decode: {}\n+ - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}\n+ - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}\n+ - Permute: {}\n+ batch_size: 4\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/eval.py",
"diff": "+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing, software\n+# distributed under the License is distributed on an \"AS IS\" BASIS,\n+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+# See the License for the specific language governing permissions and\n+# limitations under the License.\n+\n+import os\n+import sys\n+import numpy as np\n+import argparse\n+import paddle\n+from ppdet.core.workspace import load_config, merge_config\n+from ppdet.core.workspace import create\n+from ppdet.metrics import COCOMetric, VOCMetric, KeyPointTopDownCOCOEval\n+from paddleslim.auto_compression.config_helpers import load_config as load_slim_config\n+from post_process import PPYOLOEPostProcess\n+\n+\n+def argsparser():\n+ parser = argparse.ArgumentParser(description=__doc__)\n+ parser.add_argument(\n+ '--config_path',\n+ type=str,\n+ default=None,\n+ help=\"path of compression strategy config.\",\n+ required=True)\n+ parser.add_argument(\n+ '--devices',\n+ type=str,\n+ default='gpu',\n+ help=\"which device used to compress.\")\n+\n+ return parser\n+\n+\n+def reader_wrapper(reader, input_list):\n+ def gen():\n+ for data in reader:\n+ in_dict = {}\n+ if isinstance(input_list, list):\n+ for input_name in input_list:\n+ in_dict[input_name] = data[input_name]\n+ elif isinstance(input_list, dict):\n+ for input_name in input_list.keys():\n+ in_dict[input_list[input_name]] = data[input_name]\n+ yield in_dict\n+\n+ return gen\n+\n+\n+def convert_numpy_data(data, metric):\n+ data_all = {}\n+ data_all = {k: np.array(v) for k, v in data.items()}\n+ if isinstance(metric, VOCMetric):\n+ for k, v in data_all.items():\n+ if not isinstance(v[0], np.ndarray):\n+ tmp_list = []\n+ for t in v:\n+ tmp_list.append(np.array(t))\n+ data_all[k] = np.array(tmp_list)\n+ else:\n+ data_all = {k: np.array(v) for k, v in data.items()}\n+ return data_all\n+\n+\n+def eval():\n+\n+ place = paddle.CUDAPlace(0) if FLAGS.devices == 'gpu' else paddle.CPUPlace()\n+ exe = paddle.static.Executor(place)\n+\n+ val_program, feed_target_names, fetch_targets = paddle.static.load_inference_model(\n+ global_config[\"model_dir\"].rstrip('/'),\n+ exe,\n+ model_filename=global_config[\"model_filename\"],\n+ params_filename=global_config[\"params_filename\"])\n+ print('Loaded model from: {}'.format(global_config[\"model_dir\"]))\n+\n+ metric = global_config['metric']\n+ for batch_id, data in enumerate(val_loader):\n+ data_all = convert_numpy_data(data, metric)\n+ data_input = {}\n+ for k, v in data.items():\n+ if isinstance(global_config['input_list'], list):\n+ if k in global_config['input_list']:\n+ data_input[k] = np.array(v)\n+ elif isinstance(global_config['input_list'], dict):\n+ if k in global_config['input_list'].keys():\n+ data_input[global_config['input_list'][k]] = np.array(v)\n+\n+ outs = exe.run(val_program,\n+ feed=data_input,\n+ fetch_list=fetch_targets,\n+ return_numpy=False)\n+ res = {}\n+ if 'arch' in global_config and global_config['arch'] == 'PPYOLOE':\n+ postprocess = PPYOLOEPostProcess(\n+ score_threshold=0.01, nms_threshold=0.6)\n+ res = postprocess(np.array(outs[0]), data_all['scale_factor'])\n+ else:\n+ for out in outs:\n+ v = np.array(out)\n+ if len(v.shape) > 1:\n+ res['bbox'] = v\n+ else:\n+ res['bbox_num'] = v\n+ metric.update(data_all, res)\n+ if batch_id % 100 == 0:\n+ print('Eval iter:', batch_id)\n+ metric.accumulate()\n+ metric.log()\n+ metric.reset()\n+\n+\n+def main():\n+ global global_config\n+ all_config = load_slim_config(FLAGS.config_path)\n+ assert \"Global\" in all_config, \"Key 'Global' not found in config file.\"\n+ global_config = all_config[\"Global\"]\n+ reader_cfg = load_config(global_config['reader_config'])\n+\n+ dataset = reader_cfg['EvalDataset']\n+ global val_loader\n+ val_loader = create('EvalReader')(reader_cfg['EvalDataset'],\n+ reader_cfg['worker_num'],\n+ return_list=True)\n+ metric = None\n+ if reader_cfg['metric'] == 'COCO':\n+ clsid2catid = {v: k for k, v in dataset.catid2clsid.items()}\n+ anno_file = dataset.get_anno()\n+ metric = COCOMetric(\n+ anno_file=anno_file, clsid2catid=clsid2catid, IouType='bbox')\n+ elif reader_cfg['metric'] == 'VOC':\n+ metric = VOCMetric(\n+ label_list=dataset.get_label_list(),\n+ class_num=reader_cfg['num_classes'],\n+ map_type=reader_cfg['map_type'])\n+ elif reader_cfg['metric'] == 'KeyPointTopDownCOCOEval':\n+ anno_file = dataset.get_anno()\n+ metric = KeyPointTopDownCOCOEval(anno_file,\n+ len(dataset), 17, 'output_eval')\n+ else:\n+ raise ValueError(\"metric currently only supports COCO and VOC.\")\n+ global_config['metric'] = metric\n+\n+ eval()\n+\n+\n+if __name__ == '__main__':\n+ paddle.enable_static()\n+ parser = argsparser()\n+ FLAGS = parser.parse_args()\n+ assert FLAGS.devices in ['cpu', 'gpu', 'xpu', 'npu']\n+ paddle.set_device(FLAGS.devices)\n+\n+ main()\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/post_process.py",
"diff": "+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing, software\n+# distributed under the License is distributed on an \"AS IS\" BASIS,\n+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+# See the License for the specific language governing permissions and\n+# limitations under the License.\n+\n+import numpy as np\n+import cv2\n+\n+\n+def hard_nms(box_scores, iou_threshold, top_k=-1, candidate_size=200):\n+ \"\"\"\n+ Args:\n+ box_scores (N, 5): boxes in corner-form and probabilities.\n+ iou_threshold: intersection over union threshold.\n+ top_k: keep top_k results. If k <= 0, keep all the results.\n+ candidate_size: only consider the candidates with the highest scores.\n+ Returns:\n+ picked: a list of indexes of the kept boxes\n+ \"\"\"\n+ scores = box_scores[:, -1]\n+ boxes = box_scores[:, :-1]\n+ picked = []\n+ indexes = np.argsort(scores)\n+ indexes = indexes[-candidate_size:]\n+ while len(indexes) > 0:\n+ current = indexes[-1]\n+ picked.append(current)\n+ if 0 < top_k == len(picked) or len(indexes) == 1:\n+ break\n+ current_box = boxes[current, :]\n+ indexes = indexes[:-1]\n+ rest_boxes = boxes[indexes, :]\n+ iou = iou_of(\n+ rest_boxes,\n+ np.expand_dims(\n+ current_box, axis=0), )\n+ indexes = indexes[iou <= iou_threshold]\n+\n+ return box_scores[picked, :]\n+\n+\n+def iou_of(boxes0, boxes1, eps=1e-5):\n+ \"\"\"Return intersection-over-union (Jaccard index) of boxes.\n+ Args:\n+ boxes0 (N, 4): ground truth boxes.\n+ boxes1 (N or 1, 4): predicted boxes.\n+ eps: a small number to avoid 0 as denominator.\n+ Returns:\n+ iou (N): IoU values.\n+ \"\"\"\n+ overlap_left_top = np.maximum(boxes0[..., :2], boxes1[..., :2])\n+ overlap_right_bottom = np.minimum(boxes0[..., 2:], boxes1[..., 2:])\n+\n+ overlap_area = area_of(overlap_left_top, overlap_right_bottom)\n+ area0 = area_of(boxes0[..., :2], boxes0[..., 2:])\n+ area1 = area_of(boxes1[..., :2], boxes1[..., 2:])\n+ return overlap_area / (area0 + area1 - overlap_area + eps)\n+\n+\n+def area_of(left_top, right_bottom):\n+ \"\"\"Compute the areas of rectangles given two corners.\n+ Args:\n+ left_top (N, 2): left top corner.\n+ right_bottom (N, 2): right bottom corner.\n+ Returns:\n+ area (N): return the area.\n+ \"\"\"\n+ hw = np.clip(right_bottom - left_top, 0.0, None)\n+ return hw[..., 0] * hw[..., 1]\n+\n+\n+class PPYOLOEPostProcess(object):\n+ \"\"\"\n+ Args:\n+ input_shape (int): network input image size\n+ scale_factor (float): scale factor of ori image\n+ \"\"\"\n+\n+ def __init__(self,\n+ score_threshold=0.4,\n+ nms_threshold=0.5,\n+ nms_top_k=10000,\n+ keep_top_k=300):\n+ self.score_threshold = score_threshold\n+ self.nms_threshold = nms_threshold\n+ self.nms_top_k = nms_top_k\n+ self.keep_top_k = keep_top_k\n+\n+ def _non_max_suppression(self, prediction, scale_factor):\n+ batch_size = prediction.shape[0]\n+ out_boxes_list = []\n+ box_num_list = []\n+ for batch_id in range(batch_size):\n+ bboxes, confidences = prediction[batch_id][..., :4], prediction[\n+ batch_id][..., 4:]\n+ # nms\n+ picked_box_probs = []\n+ picked_labels = []\n+ for class_index in range(0, confidences.shape[1]):\n+ probs = confidences[:, class_index]\n+ mask = probs > self.score_threshold\n+ probs = probs[mask]\n+ if probs.shape[0] == 0:\n+ continue\n+ subset_boxes = bboxes[mask, :]\n+ box_probs = np.concatenate(\n+ [subset_boxes, probs.reshape(-1, 1)], axis=1)\n+ box_probs = hard_nms(\n+ box_probs,\n+ iou_threshold=self.nms_threshold,\n+ top_k=self.nms_top_k)\n+ picked_box_probs.append(box_probs)\n+ picked_labels.extend([class_index] * box_probs.shape[0])\n+\n+ if len(picked_box_probs) == 0:\n+ out_boxes_list.append(np.empty((0, 4)))\n+\n+ else:\n+ picked_box_probs = np.concatenate(picked_box_probs)\n+ # resize output boxes\n+ picked_box_probs[:, 0] /= scale_factor[batch_id][1]\n+ picked_box_probs[:, 2] /= scale_factor[batch_id][1]\n+ picked_box_probs[:, 1] /= scale_factor[batch_id][0]\n+ picked_box_probs[:, 3] /= scale_factor[batch_id][0]\n+\n+ # clas score box\n+ out_box = np.concatenate(\n+ [\n+ np.expand_dims(\n+ np.array(picked_labels), axis=-1), np.expand_dims(\n+ picked_box_probs[:, 4], axis=-1),\n+ picked_box_probs[:, :4]\n+ ],\n+ axis=1)\n+ if out_box.shape[0] > self.keep_top_k:\n+ out_box = out_box[out_box[:, 1].argsort()[::-1]\n+ [:self.keep_top_k]]\n+ out_boxes_list.append(out_box)\n+ box_num_list.append(out_box.shape[0])\n+\n+ out_boxes_list = np.concatenate(out_boxes_list, axis=0)\n+ box_num_list = np.array(box_num_list)\n+ return out_boxes_list, box_num_list\n+\n+ def __call__(self, outs, scale_factor):\n+ out_boxes_list, box_num_list = self._non_max_suppression(outs,\n+ scale_factor)\n+ return {'bbox': out_boxes_list, 'bbox_num': box_num_list}\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/run.py",
"diff": "+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.\n+#\n+# Licensed under the Apache License, Version 2.0 (the \"License\");\n+# you may not use this file except in compliance with the License.\n+# You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing, software\n+# distributed under the License is distributed on an \"AS IS\" BASIS,\n+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+# See the License for the specific language governing permissions and\n+# limitations under the License.\n+\n+import os\n+import sys\n+import numpy as np\n+import argparse\n+import paddle\n+from ppdet.core.workspace import load_config, merge_config\n+from ppdet.core.workspace import create\n+from ppdet.metrics import COCOMetric, VOCMetric, KeyPointTopDownCOCOEval\n+from paddleslim.auto_compression.config_helpers import load_config as load_slim_config\n+from paddleslim.auto_compression import AutoCompression\n+from post_process import PPYOLOEPostProcess\n+\n+\n+def argsparser():\n+ parser = argparse.ArgumentParser(description=__doc__)\n+ parser.add_argument(\n+ '--config_path',\n+ type=str,\n+ default=None,\n+ help=\"path of compression strategy config.\",\n+ required=True)\n+ parser.add_argument(\n+ '--save_dir',\n+ type=str,\n+ default='output',\n+ help=\"directory to save compressed model.\")\n+ parser.add_argument(\n+ '--devices',\n+ type=str,\n+ default='gpu',\n+ help=\"which device used to compress.\")\n+\n+ return parser\n+\n+\n+def reader_wrapper(reader, input_list):\n+ def gen():\n+ for data in reader:\n+ in_dict = {}\n+ if isinstance(input_list, list):\n+ for input_name in input_list:\n+ in_dict[input_name] = data[input_name]\n+ elif isinstance(input_list, dict):\n+ for input_name in input_list.keys():\n+ in_dict[input_list[input_name]] = data[input_name]\n+ yield in_dict\n+\n+ return gen\n+\n+\n+def convert_numpy_data(data, metric):\n+ data_all = {}\n+ data_all = {k: np.array(v) for k, v in data.items()}\n+ if isinstance(metric, VOCMetric):\n+ for k, v in data_all.items():\n+ if not isinstance(v[0], np.ndarray):\n+ tmp_list = []\n+ for t in v:\n+ tmp_list.append(np.array(t))\n+ data_all[k] = np.array(tmp_list)\n+ else:\n+ data_all = {k: np.array(v) for k, v in data.items()}\n+ return data_all\n+\n+\n+def eval_function(exe, compiled_test_program, test_feed_names, test_fetch_list):\n+ metric = global_config['metric']\n+ for batch_id, data in enumerate(val_loader):\n+ data_all = convert_numpy_data(data, metric)\n+ data_input = {}\n+ for k, v in data.items():\n+ if isinstance(global_config['input_list'], list):\n+ if k in test_feed_names:\n+ data_input[k] = np.array(v)\n+ elif isinstance(global_config['input_list'], dict):\n+ if k in global_config['input_list'].keys():\n+ data_input[global_config['input_list'][k]] = np.array(v)\n+ outs = exe.run(compiled_test_program,\n+ feed=data_input,\n+ fetch_list=test_fetch_list,\n+ return_numpy=False)\n+ res = {}\n+ if 'arch' in global_config and global_config['arch'] == 'PPYOLOE':\n+ postprocess = PPYOLOEPostProcess(\n+ score_threshold=0.01, nms_threshold=0.6)\n+ res = postprocess(np.array(outs[0]), data_all['scale_factor'])\n+ else:\n+ for out in outs:\n+ v = np.array(out)\n+ if len(v.shape) > 1:\n+ res['bbox'] = v\n+ else:\n+ res['bbox_num'] = v\n+\n+ metric.update(data_all, res)\n+ if batch_id % 100 == 0:\n+ print('Eval iter:', batch_id)\n+ metric.accumulate()\n+ metric.log()\n+ map_res = metric.get_results()\n+ metric.reset()\n+ map_key = 'keypoint' if 'arch' in global_config and global_config[\n+ 'arch'] == 'keypoint' else 'bbox'\n+ return map_res[map_key][0]\n+\n+\n+def main():\n+ global global_config\n+ all_config = load_slim_config(FLAGS.config_path)\n+ assert \"Global\" in all_config, \"Key 'Global' not found in config file.\"\n+ global_config = all_config[\"Global\"]\n+ reader_cfg = load_config(global_config['reader_config'])\n+\n+ train_loader = create('EvalReader')(reader_cfg['TrainDataset'],\n+ reader_cfg['worker_num'],\n+ return_list=True)\n+ train_loader = reader_wrapper(train_loader, global_config['input_list'])\n+\n+ if 'Evaluation' in global_config.keys() and global_config[\n+ 'Evaluation'] and paddle.distributed.get_rank() == 0:\n+ eval_func = eval_function\n+ dataset = reader_cfg['EvalDataset']\n+ global val_loader\n+ _eval_batch_sampler = paddle.io.BatchSampler(\n+ dataset, batch_size=reader_cfg['EvalReader']['batch_size'])\n+ val_loader = create('EvalReader')(dataset,\n+ reader_cfg['worker_num'],\n+ batch_sampler=_eval_batch_sampler,\n+ return_list=True)\n+ metric = None\n+ if reader_cfg['metric'] == 'COCO':\n+ clsid2catid = {v: k for k, v in dataset.catid2clsid.items()}\n+ anno_file = dataset.get_anno()\n+ metric = COCOMetric(\n+ anno_file=anno_file, clsid2catid=clsid2catid, IouType='bbox')\n+ elif reader_cfg['metric'] == 'VOC':\n+ metric = VOCMetric(\n+ label_list=dataset.get_label_list(),\n+ class_num=reader_cfg['num_classes'],\n+ map_type=reader_cfg['map_type'])\n+ elif reader_cfg['metric'] == 'KeyPointTopDownCOCOEval':\n+ anno_file = dataset.get_anno()\n+ metric = KeyPointTopDownCOCOEval(anno_file,\n+ len(dataset), 17, 'output_eval')\n+ else:\n+ raise ValueError(\"metric currently only supports COCO and VOC.\")\n+ global_config['metric'] = metric\n+ else:\n+ eval_func = None\n+\n+ ac = AutoCompression(\n+ model_dir=global_config[\"model_dir\"],\n+ model_filename=global_config[\"model_filename\"],\n+ params_filename=global_config[\"params_filename\"],\n+ save_dir=FLAGS.save_dir,\n+ config=all_config,\n+ train_dataloader=train_loader,\n+ eval_callback=eval_func)\n+ ac.compress()\n+\n+\n+if __name__ == '__main__':\n+ paddle.enable_static()\n+ parser = argsparser()\n+ FLAGS = parser.parse_args()\n+ assert FLAGS.devices in ['cpu', 'gpu', 'xpu', 'npu']\n+ paddle.set_device(FLAGS.devices)\n+\n+ main()\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add PP-YOLOE Auto Compression demo (#6568)
|
499,339 |
04.08.2022 17:31:10
| -28,800 |
befeaeb5424fcadaa70a2ff646a6a4b9c2ebf848
|
[dev] add white and black list for amp train
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -69,6 +69,8 @@ class Trainer(object):\nself.is_loaded_weights = False\nself.use_amp = self.cfg.get('amp', False)\nself.amp_level = self.cfg.get('amp_level', 'O1')\n+ self.custom_white_list = self.cfg.get('custom_white_list', None)\n+ self.custom_black_list = self.cfg.get('custom_black_list', None)\n# build data loader\ncapital_mode = self.mode.capitalize()\n@@ -155,8 +157,10 @@ class Trainer(object):\nself.pruner = create('UnstructuredPruner')(self.model,\nsteps_per_epoch)\nif self.use_amp and self.amp_level == 'O2':\n- self.model = paddle.amp.decorate(\n- models=self.model, level=self.amp_level)\n+ self.model, self.optimizer = paddle.amp.decorate(\n+ models=self.model,\n+ optimizers=self.optimizer,\n+ level=self.amp_level)\nself.use_ema = ('use_ema' in cfg and cfg['use_ema'])\nif self.use_ema:\nema_decay = self.cfg.get('ema_decay', 0.9998)\n@@ -456,7 +460,9 @@ class Trainer(object):\nDataParallel) and use_fused_allreduce_gradients:\nwith model.no_sync():\nwith paddle.amp.auto_cast(\n- enable=self.cfg.use_gpus,\n+ enable=self.cfg.use_gpu,\n+ custom_white_list=self.custom_white_list,\n+ custom_black_list=self.custom_black_list,\nlevel=self.amp_level):\n# model forward\noutputs = model(data)\n@@ -468,7 +474,10 @@ class Trainer(object):\nlist(model.parameters()), None)\nelse:\nwith paddle.amp.auto_cast(\n- enable=self.cfg.use_gpu, level=self.amp_level):\n+ enable=self.cfg.use_gpu,\n+ custom_white_list=self.custom_white_list,\n+ custom_black_list=self.custom_black_list,\n+ level=self.amp_level):\n# model forward\noutputs = model(data)\nloss = outputs['loss']\n@@ -477,7 +486,6 @@ class Trainer(object):\nscaled_loss.backward()\n# in dygraph mode, optimizer.minimize is equal to optimizer.step\nscaler.minimize(self.optimizer, scaled_loss)\n-\nelse:\nif isinstance(\nmodel, paddle.\n@@ -575,7 +583,10 @@ class Trainer(object):\n# forward\nif self.use_amp:\nwith paddle.amp.auto_cast(\n- enable=self.cfg.use_gpu, level=self.amp_level):\n+ enable=self.cfg.use_gpu,\n+ custom_white_list=self.custom_white_list,\n+ custom_black_list=self.custom_black_list,\n+ level=self.amp_level):\nouts = self.model(data)\nelse:\nouts = self.model(data)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer/ema.py",
"new_path": "ppdet/optimizer/ema.py",
"diff": "@@ -66,7 +66,10 @@ class ModelEMA(object):\ndef resume(self, state_dict, step=0):\nfor k, v in state_dict.items():\nif k in self.state_dict:\n+ if self.state_dict[k].dtype == v.dtype:\nself.state_dict[k] = v\n+ else:\n+ self.state_dict[k] = v.astype(self.state_dict[k].dtype)\nself.step = step\ndef update(self, model=None):\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/utils/checkpoint.py",
"new_path": "ppdet/utils/checkpoint.py",
"diff": "@@ -84,9 +84,14 @@ def load_weight(model, weight, optimizer=None, ema=None):\nmodel_weight = {}\nincorrect_keys = 0\n- for key in model_dict.keys():\n+ for key, value in model_dict.items():\nif key in param_state_dict.keys():\n+ if isinstance(param_state_dict[key], np.ndarray):\n+ param_state_dict[key] = paddle.to_tensor(param_state_dict[key])\n+ if value.dtype == param_state_dict[key].dtype:\nmodel_weight[key] = param_state_dict[key]\n+ else:\n+ model_weight[key] = param_state_dict[key].astype(value.dtype)\nelse:\nlogger.info('Unmatched key: {}'.format(key))\nincorrect_keys += 1\n@@ -209,6 +214,12 @@ def load_pretrain_weight(model, pretrain_weight):\nparam_state_dict = paddle.load(weights_path)\nparam_state_dict = match_state_dict(model_dict, param_state_dict)\n+ for k, v in param_state_dict.items():\n+ if isinstance(v, np.ndarray):\n+ v = paddle.to_tensor(v)\n+ if model_dict[k].dtype != v.dtype:\n+ param_state_dict[k] = v.astype(model_dict[k].dtype)\n+\nmodel.set_dict(param_state_dict)\nlogger.info('Finish loading model weights: {}'.format(weights_path))\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] add white and black list for amp train (#6576)
|
499,299 |
08.08.2022 11:07:26
| -28,800 |
09e7665d08920f4387396c65c4f1a7be2584574b
|
add attr doc for ppvehicle
|
[
{
"change_type": "ADD",
"old_path": "deploy/pipeline/docs/images/vehicle_attribute.gif",
"new_path": "deploy/pipeline/docs/images/vehicle_attribute.gif",
"diff": "Binary files /dev/null and b/deploy/pipeline/docs/images/vehicle_attribute.gif differ\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -42,7 +42,7 @@ from utils import argsparser, Timer, get_current_memory_mb\nSUPPORT_MODELS = {\n'YOLO', 'RCNN', 'SSD', 'Face', 'FCOS', 'SOLOv2', 'TTFNet', 'S2ANet', 'JDE',\n'FairMOT', 'DeepSORT', 'GFL', 'PicoDet', 'CenterNet', 'TOOD', 'RetinaNet',\n- 'StrongBaseline', 'STGCN', 'YOLOX', 'PPHGNet'\n+ 'StrongBaseline', 'STGCN', 'YOLOX', 'PPHGNet', 'PPLCNet'\n}\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add attr doc for ppvehicle (#6593)
|
499,339 |
08.08.2022 16:24:26
| -28,800 |
494f381f213b3ca863e74048d193e246916b767b
|
[TIPC] fix random seed in train benchmark
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/keypoint/tinypose_128x96_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/keypoint/tinypose_128x96_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/keypoint/tinypose_128x96.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c test_tipc/configs/keypoint/tinypose_128x96.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_1x_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/picodet/picodet_lcnet_1_5x_416_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/picodet/picodet_lcnet_1_5x_416_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/picodet_lcnet_1_5x_416_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/picodet/legacy_model/more_config/picodet_lcnet_1_5x_416_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/picodet/picodet_s_320_coco_lcnet_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/picodet/picodet_s_320_coco_lcnet_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/picodet_s_320_coco_lcnet.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/picodet/picodet_s_320_coco_lcnet.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/picodet/picodet_s_320_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/picodet/picodet_s_320_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/picodet_s_320_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/picodet/legacy_model/picodet_s_320_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyolo/ppyolo_mbv3_large_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/ppyolo/ppyolo_mbv3_large_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyolo/ppyolo_tiny_650e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/ppyolo/ppyolo_tiny_650e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolo_tiny_650e_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/ppyolo/ppyolo_tiny_650e_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/ppyolov2_r50vd_dcn_365e_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/ppyoloe/ppyoloe_crn_s_300e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/ppyoloe/ppyoloe_crn_s_300e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/ppyoloe_crn_s_300e_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/yolov3/yolov3_darknet53_270e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"new_path": "test_tipc/configs/yolov3/yolov3_darknet53_270e_coco_train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt",
"diff": "@@ -10,7 +10,7 @@ TrainReader.batch_size:lite_train_lite_infer=2|lite_train_whole_infer=2|whole_tr\npretrain_weights:https://paddledet.bj.bcebos.com/models/yolov3_darknet53_270e_coco.pdparams\ntrained_model_name:model_final.pdparams\ntrain_infer_img_dir:./dataset/coco/test2017/\n-amp_level:O2\n+null:null\n##\ntrainer:norm_train\nnorm_train:tools/train.py -c configs/yolov3/yolov3_darknet53_270e_coco.yml -o\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_train_inference_python.sh",
"new_path": "test_tipc/test_train_inference_python.sh",
"diff": "@@ -271,17 +271,25 @@ else\nsave_log=\"${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}\"\nif [ ${autocast} = \"amp\" ] || [ ${autocast} = \"fp16\" ]; then\nset_autocast=\"--amp\"\n- set_train_params1=\"amp_level=O2\"\n+ set_amp_level=\"amp_level=O2\"\nelse\nset_autocast=\" \"\n+ set_amp_level=\" \"\n+ fi\n+ if [ ${MODE} = \"benchmark_train\" ]; then\n+ set_shuffle=\"TrainReader.shuffle=False\"\n+ set_enable_ce=\"--enable_ce=True\"\n+ else\n+ set_shuffle=\" \"\n+ set_enable_ce=\" \"\nfi\nset_save_model=$(func_set_params \"${save_model_key}\" \"${save_log}\")\nnodes=\"1\"\nif [ ${#gpu} -le 2 ];then # train with cpu or single gpu\n- cmd=\"${python} ${run_train} LearningRate.base_lr=0.0001 log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_train_params1} ${set_autocast}\"\n+ cmd=\"${python} ${run_train} LearningRate.base_lr=0.0001 log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_shuffle} ${set_amp_level} ${set_enable_ce} ${set_autocast} ${set_train_params1}\"\nelif [ ${#ips} -le 15 ];then # train with multi-gpu\n- cmd=\"${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train} log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_train_params1} ${set_autocast}\"\n+ cmd=\"${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train} log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_shuffle} ${set_amp_level} ${set_enable_ce} ${set_autocast} ${set_train_params1}\"\nelse # train with multi-machine\nIFS=\",\"\nips_array=(${ips})\n@@ -289,7 +297,7 @@ else\nsave_log=\"${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}\"\nIFS=\"|\"\nset_save_model=$(func_set_params \"${save_model_key}\" \"${save_log}\")\n- cmd=\"${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_train_params1} ${set_autocast}\"\n+ cmd=\"${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} log_iter=1 ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_filename} ${set_shuffle} ${set_amp_level} ${set_enable_ce} ${set_autocast} ${set_train_params1}\"\nfi\n# run train\ntrain_log_path=\"${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}.log\"\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix random seed in train benchmark (#6603)
|
499,339 |
10.08.2022 11:12:44
| -28,800 |
cb89c8d0567073aa3a1506fc679617c8ac13a1b2
|
[dev] fix trt nms error output in ppyoloe
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/layers.py",
"new_path": "ppdet/modeling/layers.py",
"diff": "@@ -481,8 +481,9 @@ class MultiClassNMS(object):\n# TODO(wangxinxin08): tricky switch to run nms on tensorrt\nkwargs.update({'nms_eta': 1.1})\nbbox, bbox_num, _ = ops.multiclass_nms(bboxes, score, **kwargs)\n- mask = paddle.slice(bbox, [-1], [0], [1]) != -1\n- bbox = paddle.masked_select(bbox, mask).reshape((-1, 6))\n+ bbox = bbox.reshape([1, -1, 6])\n+ idx = paddle.nonzero(bbox[..., 0] != -1)\n+ bbox = paddle.gather_nd(bbox, idx)\nreturn bbox, bbox_num, None\nelse:\nreturn ops.multiclass_nms(bboxes, score, **kwargs)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] fix trt nms error output in ppyoloe (#6607)
|
499,333 |
10.08.2022 21:16:04
| -28,800 |
42a4d70710f019ab3219308baf17937e5cf5ec34
|
update qq qr-code, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "README_en.md",
"new_path": "README_en.md",
"diff": "Welcome to join PaddleDetection user groups on QQ, WeChat (scan the QR code, add and reply \"D\" to the assistant)\n<div align=\"center\">\n- <img src=\"https://user-images.githubusercontent.com/48054808/157800129-2f9a0b72-6bb8-4b10-8310-93ab1639253f.jpg\" width = \"200\" />\n- <img src=\"https://user-images.githubusercontent.com/48054808/160531099-9811bbe6-cfbb-47d5-8bdb-c2b40684d7dd.png\" width = \"200\" />\n+ <img src=\"https://user-images.githubusercontent.com/22989727/183843004-baebf75f-af7c-4a7c-8130-1497b9a3ec7e.png\" width = \"200\" />\n+ <img src=\"https://user-images.githubusercontent.com/34162360/177678712-4655747d-4290-4ad9-b7a1-4564a5418ac6.jpg\" width = \"200\" />\n</div>\n## <img src=\"https://user-images.githubusercontent.com/48054808/157827140-03ffaff7-7d14-48b4-9440-c38986ea378c.png\" width=\"20\"/> Kit Structure\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update qq qr-code, test=document_fix (#6626)
|
499,301 |
11.08.2022 17:05:50
| -28,800 |
2607dbca4b643b79c66b245152d491e4afee0327
|
recompute flag
|
[
{
"change_type": "MODIFY",
"old_path": "configs/vitdet/cascade_rcnn_vit_large_hrfpn_cae_1x_coco.yml",
"new_path": "configs/vitdet/cascade_rcnn_vit_large_hrfpn_cae_1x_coco.yml",
"diff": "@@ -7,6 +7,7 @@ weights: output/cascade_rcnn_vit_large_hrfpn_cae_1x_coco/model_final\ndepth: &depth 24\ndim: &dim 1024\n+use_fused_allreduce_gradients: &use_checkpoint True\nVisionTransformer:\nimg_size: [800, 1344]\n@@ -15,6 +16,7 @@ VisionTransformer:\nnum_heads: 16\ndrop_path_rate: 0.25\nout_indices: [7, 11, 15, 23]\n+ use_checkpoint: *use_checkpoint\npretrained: https://bj.bcebos.com/v1/paddledet/models/pretrained/vit_large_cae_pretrained.pdparams\nHRFPN:\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/vision_transformer.py",
"new_path": "ppdet/modeling/backbones/vision_transformer.py",
"diff": "@@ -596,7 +596,7 @@ class VisionTransformer(nn.Layer):\nfeats = []\nfor idx, blk in enumerate(self.blocks):\n- if self.use_checkpoint:\n+ if self.use_checkpoint and self.training:\nx = paddle.distributed.fleet.utils.recompute(\nblk, x, rel_pos_bias, **{\"preserve_rng_state\": True})\nelse:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
recompute flag (#6628)
|
499,298 |
11.08.2022 17:22:26
| -28,800 |
936ec224e9cb64413e962c37dfa5fb5e7f37d31e
|
add Resize export default interp
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/export_utils.py",
"new_path": "ppdet/engine/export_utils.py",
"diff": "@@ -92,6 +92,7 @@ def _parse_reader(reader_cfg, dataset_cfg, metric, arch, image_shape):\nif key == 'Resize':\nif int(image_shape[1]) != -1:\nvalue['target_size'] = image_shape[1:]\n+ value['interp'] = value.get('interp', 1) # cv2.INTER_LINEAR\nif fuse_normalize and key == 'NormalizeImage':\ncontinue\np.update(value)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add Resize export default interp (#6632)
|
499,299 |
15.08.2022 10:40:43
| -28,800 |
ff8a7b1d090a2f57048d3e87892706a8407dcfe6
|
move initialize part into class
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipeline.py",
"new_path": "deploy/pipeline/pipeline.py",
"diff": "@@ -60,29 +60,8 @@ class Pipeline(object):\nPipeline\nArgs:\n+ args (argparse.Namespace): arguments in pipeline, which contains environment and runtime settings\ncfg (dict): config of models in pipeline\n- image_file (string|None): the path of image file, default as None\n- image_dir (string|None): the path of image directory, if not None,\n- then all the images in directory will be predicted, default as None\n- video_file (string|None): the path of video file, default as None\n- camera_id (int): the device id of camera to predict, default as -1\n- device (string): the device to predict, options are: CPU/GPU/XPU,\n- default as CPU\n- run_mode (string): the mode of prediction, options are:\n- paddle/trt_fp32/trt_fp16, default as paddle\n- trt_min_shape (int): min shape for dynamic shape in trt, default as 1\n- trt_max_shape (int): max shape for dynamic shape in trt, default as 1280\n- trt_opt_shape (int): opt shape for dynamic shape in trt, default as 640\n- trt_calib_mode (bool): If the model is produced by TRT offline quantitative\n- calibration, trt_calib_mode need to set True. default as False\n- cpu_threads (int): cpu threads, default as 1\n- enable_mkldnn (bool): whether to open MKLDNN, default as False\n- output_dir (string): The path of output, default as 'output'\n- draw_center_traj (bool): Whether drawing the trajectory of center, default as False\n- secs_interval (int): The seconds interval to count after tracking, default as 10\n- do_entrance_counting(bool): Whether counting the numbers of identifiers entering\n- or getting out from the entrance, default as False, only support single class\n- counting in MOT.\n\"\"\"\ndef __init__(self, args, cfg):\n@@ -108,18 +87,6 @@ class Pipeline(object):\nif self.is_video:\nself.predictor.set_file_name(args.video_file)\n- self.output_dir = args.output_dir\n- self.draw_center_traj = args.draw_center_traj\n- self.secs_interval = args.secs_interval\n- self.do_entrance_counting = args.do_entrance_counting\n- self.do_break_in_counting = args.do_break_in_counting\n- self.region_type = args.region_type\n- self.region_polygon = args.region_polygon\n- if self.region_type == 'custom':\n- assert len(\n- self.region_polygon\n- ) > 6, 'region_type is custom, region_polygon should be at least 3 pairs of point coords.'\n-\ndef _parse_input(self, image_file, image_dir, video_file, video_dir,\ncamera_id):\n@@ -179,8 +146,10 @@ class Pipeline(object):\ndef get_model_dir(cfg):\n- # auto download inference model\n- model_dir_dict = {}\n+ \"\"\"\n+ Auto download inference model if the model_path is a url link.\n+ Otherwise it will use the model_path directly.\n+ \"\"\"\nfor key in cfg.keys():\nif type(cfg[key]) == dict and \\\n(\"enable\" in cfg[key].keys() and cfg[key]['enable']\n@@ -191,30 +160,30 @@ def get_model_dir(cfg):\ndownloaded_model_dir = auto_download_model(model_dir)\nif downloaded_model_dir:\nmodel_dir = downloaded_model_dir\n- model_dir_dict[key] = model_dir\n+ cfg[key][\"model_dir\"] = model_dir\nprint(key, \" model dir: \", model_dir)\nelif key == \"VEHICLE_PLATE\":\ndet_model_dir = cfg[key][\"det_model_dir\"]\ndownloaded_det_model_dir = auto_download_model(det_model_dir)\nif downloaded_det_model_dir:\ndet_model_dir = downloaded_det_model_dir\n- model_dir_dict[\"det_model_dir\"] = det_model_dir\n+ cfg[key][\"det_model_dir\"] = det_model_dir\nprint(\"det_model_dir model dir: \", det_model_dir)\nrec_model_dir = cfg[key][\"rec_model_dir\"]\ndownloaded_rec_model_dir = auto_download_model(rec_model_dir)\nif downloaded_rec_model_dir:\nrec_model_dir = downloaded_rec_model_dir\n- model_dir_dict[\"rec_model_dir\"] = rec_model_dir\n+ cfg[key][\"rec_model_dir\"] = rec_model_dir\nprint(\"rec_model_dir model dir: \", rec_model_dir)\n+\nelif key == \"MOT\": # for idbased and skeletonbased actions\nmodel_dir = cfg[key][\"model_dir\"]\ndownloaded_model_dir = auto_download_model(model_dir)\nif downloaded_model_dir:\nmodel_dir = downloaded_model_dir\n- model_dir_dict[key] = model_dir\n-\n- return model_dir_dict\n+ cfg[key][\"model_dir\"] = model_dir\n+ print(\"mot_model_dir model_dir: \", model_dir)\nclass PipePredictor(object):\n@@ -234,47 +203,14 @@ class PipePredictor(object):\n4. VideoAction Recognition\nArgs:\n+ args (argparse.Namespace): arguments in pipeline, which contains environment and runtime settings\ncfg (dict): config of models in pipeline\nis_video (bool): whether the input is video, default as False\nmulti_camera (bool): whether to use multi camera in pipeline,\ndefault as False\n- camera_id (int): the device id of camera to predict, default as -1\n- device (string): the device to predict, options are: CPU/GPU/XPU,\n- default as CPU\n- run_mode (string): the mode of prediction, options are:\n- paddle/trt_fp32/trt_fp16, default as paddle\n- trt_min_shape (int): min shape for dynamic shape in trt, default as 1\n- trt_max_shape (int): max shape for dynamic shape in trt, default as 1280\n- trt_opt_shape (int): opt shape for dynamic shape in trt, default as 640\n- trt_calib_mode (bool): If the model is produced by TRT offline quantitative\n- calibration, trt_calib_mode need to set True. default as False\n- cpu_threads (int): cpu threads, default as 1\n- enable_mkldnn (bool): whether to open MKLDNN, default as False\n- output_dir (string): The path of output, default as 'output'\n- draw_center_traj (bool): Whether drawing the trajectory of center, default as False\n- secs_interval (int): The seconds interval to count after tracking, default as 10\n- do_entrance_counting(bool): Whether counting the numbers of identifiers entering\n- or getting out from the entrance, default as False, only support single class\n- counting in MOT.\n\"\"\"\ndef __init__(self, args, cfg, is_video=True, multi_camera=False):\n- device = args.device\n- run_mode = args.run_mode\n- trt_min_shape = args.trt_min_shape\n- trt_max_shape = args.trt_max_shape\n- trt_opt_shape = args.trt_opt_shape\n- trt_calib_mode = args.trt_calib_mode\n- cpu_threads = args.cpu_threads\n- enable_mkldnn = args.enable_mkldnn\n- output_dir = args.output_dir\n- draw_center_traj = args.draw_center_traj\n- secs_interval = args.secs_interval\n- do_entrance_counting = args.do_entrance_counting\n- do_break_in_counting = args.do_break_in_counting\n- region_type = args.region_type\n- region_polygon = args.region_polygon\n-\n# general module for pphuman and ppvehicle\nself.with_mot = cfg.get('MOT', False)['enable'] if cfg.get(\n'MOT', False) else False\n@@ -347,13 +283,13 @@ class PipePredictor(object):\nself.is_video = is_video\nself.multi_camera = multi_camera\nself.cfg = cfg\n- self.output_dir = output_dir\n- self.draw_center_traj = draw_center_traj\n- self.secs_interval = secs_interval\n- self.do_entrance_counting = do_entrance_counting\n- self.do_break_in_counting = do_break_in_counting\n- self.region_type = region_type\n- self.region_polygon = region_polygon\n+ self.output_dir = args.output_dir\n+ self.draw_center_traj = args.draw_center_traj\n+ self.secs_interval = args.secs_interval\n+ self.do_entrance_counting = args.do_entrance_counting\n+ self.do_break_in_counting = args.do_break_in_counting\n+ self.region_type = args.region_type\n+ self.region_polygon = args.region_polygon\nself.warmup_frame = self.cfg['warmup_frame']\nself.pipeline_res = Result()\n@@ -362,7 +298,7 @@ class PipePredictor(object):\nself.collector = DataCollector()\n# auto download inference model\n- model_dir_dict = get_model_dir(self.cfg)\n+ get_model_dir(self.cfg)\nif self.with_vehicleplate:\nvehicleplate_cfg = self.cfg['VEHICLE_PLATE']\n@@ -372,148 +308,84 @@ class PipePredictor(object):\nif self.with_human_attr:\nattr_cfg = self.cfg['ATTR']\n- model_dir = model_dir_dict['ATTR']\n- batch_size = attr_cfg['batch_size']\nbasemode = self.basemode['ATTR']\nself.modebase[basemode] = True\n- self.attr_predictor = AttrDetector(\n- model_dir, device, run_mode, batch_size, trt_min_shape,\n- trt_max_shape, trt_opt_shape, trt_calib_mode, cpu_threads,\n- enable_mkldnn)\n+ self.attr_predictor = AttrDetector.init_with_cfg(args, attr_cfg)\nif self.with_vehicle_attr:\nvehicleattr_cfg = self.cfg['VEHICLE_ATTR']\n- model_dir = model_dir_dict['VEHICLE_ATTR']\n- batch_size = vehicleattr_cfg['batch_size']\n- color_threshold = vehicleattr_cfg['color_threshold']\n- type_threshold = vehicleattr_cfg['type_threshold']\nbasemode = self.basemode['VEHICLE_ATTR']\nself.modebase[basemode] = True\n- self.vehicle_attr_predictor = VehicleAttr(\n- model_dir, device, run_mode, batch_size, trt_min_shape,\n- trt_max_shape, trt_opt_shape, trt_calib_mode, cpu_threads,\n- enable_mkldnn, color_threshold, type_threshold)\n+ self.vehicle_attr_predictor = VehicleAttr.init_with_cfg(\n+ args, vehicleattr_cfg)\nif not is_video:\ndet_cfg = self.cfg['DET']\n- model_dir = model_dir_dict['DET']\n+ model_dir = det_cfg['model_dir']\nbatch_size = det_cfg['batch_size']\nself.det_predictor = Detector(\n- model_dir, device, run_mode, batch_size, trt_min_shape,\n- trt_max_shape, trt_opt_shape, trt_calib_mode, cpu_threads,\n- enable_mkldnn)\n-\n+ model_dir, args.device, args.run_mode, batch_size,\n+ args.trt_min_shape, args.trt_max_shape, args.trt_opt_shape,\n+ args.trt_calib_mode, args.cpu_threads, args.enable_mkldnn)\nelse:\nif self.with_idbased_detaction:\nidbased_detaction_cfg = self.cfg['ID_BASED_DETACTION']\n- model_dir = model_dir_dict['ID_BASED_DETACTION']\n- batch_size = idbased_detaction_cfg['batch_size']\nbasemode = self.basemode['ID_BASED_DETACTION']\n- threshold = idbased_detaction_cfg['threshold']\n- display_frames = idbased_detaction_cfg['display_frames']\n- skip_frame_num = idbased_detaction_cfg['skip_frame_num']\nself.modebase[basemode] = True\n- self.det_action_predictor = DetActionRecognizer(\n- model_dir,\n- device,\n- run_mode,\n- batch_size,\n- trt_min_shape,\n- trt_max_shape,\n- trt_opt_shape,\n- trt_calib_mode,\n- cpu_threads,\n- enable_mkldnn,\n- threshold=threshold,\n- display_frames=display_frames,\n- skip_frame_num=skip_frame_num)\n+ self.det_action_predictor = DetActionRecognizer.init_with_cfg(\n+ args, idbased_detaction_cfg)\nself.det_action_visual_helper = ActionVisualHelper(1)\nif self.with_idbased_clsaction:\nidbased_clsaction_cfg = self.cfg['ID_BASED_CLSACTION']\n- model_dir = model_dir_dict['ID_BASED_CLSACTION']\n- batch_size = idbased_clsaction_cfg['batch_size']\nbasemode = self.basemode['ID_BASED_CLSACTION']\n- threshold = idbased_clsaction_cfg['threshold']\nself.modebase[basemode] = True\n- display_frames = idbased_clsaction_cfg['display_frames']\n- skip_frame_num = idbased_clsaction_cfg['skip_frame_num']\n- self.cls_action_predictor = ClsActionRecognizer(\n- model_dir,\n- device,\n- run_mode,\n- batch_size,\n- trt_min_shape,\n- trt_max_shape,\n- trt_opt_shape,\n- trt_calib_mode,\n- cpu_threads,\n- enable_mkldnn,\n- threshold=threshold,\n- display_frames=display_frames,\n- skip_frame_num=skip_frame_num)\n+ self.cls_action_predictor = ClsActionRecognizer.init_with_cfg(\n+ args, idbased_clsaction_cfg)\nself.cls_action_visual_helper = ActionVisualHelper(1)\nif self.with_skeleton_action:\nskeleton_action_cfg = self.cfg['SKELETON_ACTION']\n- skeleton_action_model_dir = model_dir_dict['SKELETON_ACTION']\n- skeleton_action_batch_size = skeleton_action_cfg['batch_size']\n- skeleton_action_frames = skeleton_action_cfg['max_frames']\ndisplay_frames = skeleton_action_cfg['display_frames']\nself.coord_size = skeleton_action_cfg['coord_size']\nbasemode = self.basemode['SKELETON_ACTION']\nself.modebase[basemode] = True\n+ skeleton_action_frames = skeleton_action_cfg['max_frames']\n- self.skeleton_action_predictor = SkeletonActionRecognizer(\n- skeleton_action_model_dir,\n- device,\n- run_mode,\n- skeleton_action_batch_size,\n- trt_min_shape,\n- trt_max_shape,\n- trt_opt_shape,\n- trt_calib_mode,\n- cpu_threads,\n- enable_mkldnn,\n- window_size=skeleton_action_frames)\n+ self.skeleton_action_predictor = SkeletonActionRecognizer.init_with_cfg(\n+ args, skeleton_action_cfg)\nself.skeleton_action_visual_helper = ActionVisualHelper(\ndisplay_frames)\n- if self.modebase[\"skeletonbased\"]:\nkpt_cfg = self.cfg['KPT']\n- kpt_model_dir = model_dir_dict['KPT']\n+ kpt_model_dir = kpt_cfg['model_dir']\nkpt_batch_size = kpt_cfg['batch_size']\nself.kpt_predictor = KeyPointDetector(\nkpt_model_dir,\n- device,\n- run_mode,\n+ args.device,\n+ args.run_mode,\nkpt_batch_size,\n- trt_min_shape,\n- trt_max_shape,\n- trt_opt_shape,\n- trt_calib_mode,\n- cpu_threads,\n- enable_mkldnn,\n+ args.trt_min_shape,\n+ args.trt_max_shape,\n+ args.trt_opt_shape,\n+ args.trt_calib_mode,\n+ args.cpu_threads,\n+ args.enable_mkldnn,\nuse_dark=False)\nself.kpt_buff = KeyPointBuff(skeleton_action_frames)\nif self.with_mtmct:\nreid_cfg = self.cfg['REID']\n- model_dir = model_dir_dict['REID']\n- batch_size = reid_cfg['batch_size']\nbasemode = self.basemode['REID']\nself.modebase[basemode] = True\n- self.reid_predictor = ReID(\n- model_dir, device, run_mode, batch_size, trt_min_shape,\n- trt_max_shape, trt_opt_shape, trt_calib_mode, cpu_threads,\n- enable_mkldnn)\n+ self.reid_predictor = ReID.init_with_cfg(args, reid_cfg)\nif self.with_mot or self.modebase[\"idbased\"] or self.modebase[\n\"skeletonbased\"]:\nmot_cfg = self.cfg['MOT']\n- model_dir = model_dir_dict['MOT']\n+ model_dir = mot_cfg['model_dir']\ntracker_config = mot_cfg['tracker_config']\nbatch_size = mot_cfg['batch_size']\nbasemode = self.basemode['MOT']\n@@ -521,46 +393,28 @@ class PipePredictor(object):\nself.mot_predictor = SDE_Detector(\nmodel_dir,\ntracker_config,\n- device,\n- run_mode,\n+ args.device,\n+ args.run_mode,\nbatch_size,\n- trt_min_shape,\n- trt_max_shape,\n- trt_opt_shape,\n- trt_calib_mode,\n- cpu_threads,\n- enable_mkldnn,\n- draw_center_traj=draw_center_traj,\n- secs_interval=secs_interval,\n- do_entrance_counting=do_entrance_counting,\n- do_break_in_counting=do_break_in_counting,\n- region_type=region_type,\n- region_polygon=region_polygon)\n+ args.trt_min_shape,\n+ args.trt_max_shape,\n+ args.trt_opt_shape,\n+ args.trt_calib_mode,\n+ args.cpu_threads,\n+ args.enable_mkldnn,\n+ draw_center_traj=self.draw_center_traj,\n+ secs_interval=self.secs_interval,\n+ do_entrance_counting=self.do_entrance_counting,\n+ do_break_in_counting=self.do_break_in_counting,\n+ region_type=self.region_type,\n+ region_polygon=self.region_polygon)\nif self.with_video_action:\nvideo_action_cfg = self.cfg['VIDEO_ACTION']\n-\nbasemode = self.basemode['VIDEO_ACTION']\nself.modebase[basemode] = True\n-\n- video_action_model_dir = model_dir_dict['VIDEO_ACTION']\n- video_action_batch_size = video_action_cfg['batch_size']\n- short_size = video_action_cfg[\"short_size\"]\n- target_size = video_action_cfg[\"target_size\"]\n-\n- self.video_action_predictor = VideoActionRecognizer(\n- model_dir=video_action_model_dir,\n- short_size=short_size,\n- target_size=target_size,\n- device=device,\n- run_mode=run_mode,\n- batch_size=video_action_batch_size,\n- trt_min_shape=trt_min_shape,\n- trt_max_shape=trt_max_shape,\n- trt_opt_shape=trt_opt_shape,\n- trt_calib_mode=trt_calib_mode,\n- cpu_threads=cpu_threads,\n- enable_mkldnn=enable_mkldnn)\n+ self.video_action_predictor = VideoActionRecognizer.init_with_cfg(\n+ args, video_action_cfg)\ndef set_file_name(self, path):\nif path is not None:\n@@ -701,6 +555,10 @@ class PipePredictor(object):\nassert len(\nself.region_polygon\n) % 2 == 0, \"region_polygon should be pairs of coords points when do break_in counting.\"\n+ assert len(\n+ self.region_polygon\n+ ) > 6, 'region_type is custom, region_polygon should be at least 3 pairs of point coords.'\n+\nfor i in range(0, len(self.region_polygon), 2):\nentrance.append(\n[self.region_polygon[i], self.region_polygon[i + 1]])\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pphuman/action_infer.py",
"new_path": "deploy/pipeline/pphuman/action_infer.py",
"diff": "@@ -84,6 +84,20 @@ class SkeletonActionRecognizer(Detector):\nthreshold=threshold,\ndelete_shuffle_pass=True)\n+ @classmethod\n+ def init_with_cfg(cls, args, cfg):\n+ return cls(model_dir=cfg['model_dir'],\n+ batch_size=cfg['batch_size'],\n+ window_size=cfg['max_frames'],\n+ device=args.device,\n+ run_mode=args.run_mode,\n+ trt_min_shape=args.trt_min_shape,\n+ trt_max_shape=args.trt_max_shape,\n+ trt_opt_shape=args.trt_opt_shape,\n+ trt_calib_mode=args.trt_calib_mode,\n+ cpu_threads=args.cpu_threads,\n+ enable_mkldnn=args.enable_mkldnn)\n+\ndef predict(self, repeats=1):\n'''\nArgs:\n@@ -322,6 +336,22 @@ class DetActionRecognizer(object):\nself.skip_frame_cnt = 0\nself.id_in_last_frame = []\n+ @classmethod\n+ def init_with_cfg(cls, args, cfg):\n+ return cls(model_dir=cfg['model_dir'],\n+ batch_size=cfg['batch_size'],\n+ threshold=cfg['threshold'],\n+ display_frames=cfg['display_frames'],\n+ skip_frame_num=cfg['skip_frame_num'],\n+ device=args.device,\n+ run_mode=args.run_mode,\n+ trt_min_shape=args.trt_min_shape,\n+ trt_max_shape=args.trt_max_shape,\n+ trt_opt_shape=args.trt_opt_shape,\n+ trt_calib_mode=args.trt_calib_mode,\n+ cpu_threads=args.cpu_threads,\n+ enable_mkldnn=args.enable_mkldnn)\n+\ndef predict(self, images, mot_result):\nif self.skip_frame_cnt == 0 or (not self.check_id_is_same(mot_result)):\ndet_result = self.detector.predict_image(images, visual=False)\n@@ -473,6 +503,22 @@ class ClsActionRecognizer(AttrDetector):\nself.skip_frame_cnt = 0\nself.id_in_last_frame = []\n+ @classmethod\n+ def init_with_cfg(cls, args, cfg):\n+ return cls(model_dir=cfg['model_dir'],\n+ batch_size=cfg['batch_size'],\n+ threshold=cfg['threshold'],\n+ display_frames=cfg['display_frames'],\n+ skip_frame_num=cfg['skip_frame_num'],\n+ device=args.device,\n+ run_mode=args.run_mode,\n+ trt_min_shape=args.trt_min_shape,\n+ trt_max_shape=args.trt_max_shape,\n+ trt_opt_shape=args.trt_opt_shape,\n+ trt_calib_mode=args.trt_calib_mode,\n+ cpu_threads=args.cpu_threads,\n+ enable_mkldnn=args.enable_mkldnn)\n+\ndef predict_with_mot(self, images, mot_result):\nif self.skip_frame_cnt == 0 or (not self.check_id_is_same(mot_result)):\nimages = self.crop_half_body(images)\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pphuman/attr_infer.py",
"new_path": "deploy/pipeline/pphuman/attr_infer.py",
"diff": "@@ -84,6 +84,19 @@ class AttrDetector(Detector):\noutput_dir=output_dir,\nthreshold=threshold, )\n+ @classmethod\n+ def init_with_cfg(cls, args, cfg):\n+ return cls(model_dir=cfg['model_dir'],\n+ batch_size=cfg['batch_size'],\n+ device=args.device,\n+ run_mode=args.run_mode,\n+ trt_min_shape=args.trt_min_shape,\n+ trt_max_shape=args.trt_max_shape,\n+ trt_opt_shape=args.trt_opt_shape,\n+ trt_calib_mode=args.trt_calib_mode,\n+ cpu_threads=args.cpu_threads,\n+ enable_mkldnn=args.enable_mkldnn)\n+\ndef get_label(self):\nreturn self.pred_config.labels\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pphuman/reid.py",
"new_path": "deploy/pipeline/pphuman/reid.py",
"diff": "@@ -75,6 +75,19 @@ class ReID(object):\nself.batch_size = batch_size\nself.input_wh = (128, 256)\n+ @classmethod\n+ def init_with_cfg(cls, args, cfg):\n+ return cls(model_dir=cfg['model_dir'],\n+ batch_size=cfg['batch_size'],\n+ device=args.device,\n+ run_mode=args.run_mode,\n+ trt_min_shape=args.trt_min_shape,\n+ trt_max_shape=args.trt_max_shape,\n+ trt_opt_shape=args.trt_opt_shape,\n+ trt_calib_mode=args.trt_calib_mode,\n+ cpu_threads=args.cpu_threads,\n+ enable_mkldnn=args.enable_mkldnn)\n+\ndef set_config(self, model_dir):\nreturn PredictConfig(model_dir)\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pphuman/video_action_infer.py",
"new_path": "deploy/pipeline/pphuman/video_action_infer.py",
"diff": "@@ -126,6 +126,21 @@ class VideoActionRecognizer(object):\nself.predictor = create_predictor(self.config)\n+ @classmethod\n+ def init_with_cfg(cls, args, cfg):\n+ return cls(model_dir=cfg['model_dir'],\n+ short_size=cfg['short_size'],\n+ target_size=cfg['target_size'],\n+ batch_size=cfg['batch_size'],\n+ device=args.device,\n+ run_mode=args.run_mode,\n+ trt_min_shape=args.trt_min_shape,\n+ trt_max_shape=args.trt_max_shape,\n+ trt_opt_shape=args.trt_opt_shape,\n+ trt_calib_mode=args.trt_calib_mode,\n+ cpu_threads=args.cpu_threads,\n+ enable_mkldnn=args.enable_mkldnn)\n+\ndef preprocess_batch(self, file_list):\nbatched_inputs = []\nfor file in file_list:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/ppvehicle/vehicle_attr.py",
"new_path": "deploy/pipeline/ppvehicle/vehicle_attr.py",
"diff": "@@ -90,6 +90,21 @@ class VehicleAttr(AttrDetector):\n\"estate\"\n]\n+ @classmethod\n+ def init_with_cfg(cls, args, cfg):\n+ return cls(model_dir=cfg['model_dir'],\n+ batch_size=cfg['batch_size'],\n+ color_threshold=cfg['color_threshold'],\n+ type_threshold=cfg['type_threshold'],\n+ device=args.device,\n+ run_mode=args.run_mode,\n+ trt_min_shape=args.trt_min_shape,\n+ trt_max_shape=args.trt_max_shape,\n+ trt_opt_shape=args.trt_opt_shape,\n+ trt_calib_mode=args.trt_calib_mode,\n+ cpu_threads=args.cpu_threads,\n+ enable_mkldnn=args.enable_mkldnn)\n+\ndef postprocess(self, inputs, result):\n# postprocess output of predictor\nim_results = result['output']\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
move initialize part into class (#6621)
|
499,299 |
15.08.2022 20:09:34
| -28,800 |
7f884da6fafdf4033b8039c7fc782e88375e6467
|
use link for vehicle model
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/config/examples/infer_cfg_human_attr.yml",
"new_path": "deploy/pipeline/config/examples/infer_cfg_human_attr.yml",
"diff": "@@ -3,6 +3,10 @@ attr_thresh: 0.5\nvisual: True\nwarmup_frame: 50\n+DET:\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip\n+ batch_size: 1\n+\nMOT:\nmodel_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip\ntracker_config: deploy/pipeline/config/tracker_config.yml\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/pipeline/config/examples/infer_cfg_vehicle_attr.yml",
"diff": "+crop_thresh: 0.5\n+visual: True\n+warmup_frame: 50\n+\n+DET:\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip\n+ batch_size: 1\n+\n+MOT:\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip\n+ tracker_config: deploy/pipeline/config/tracker_config.yml\n+ batch_size: 1\n+ enable: True\n+\n+VEHICLE_ATTR:\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/vehicle_attribute_model.zip\n+ batch_size: 8\n+ color_threshold: 0.5\n+ type_threshold: 0.5\n+ enable: True\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/pipeline/config/examples/infer_cfg_vehicle_plate.yml",
"diff": "+crop_thresh: 0.5\n+visual: True\n+warmup_frame: 50\n+\n+DET:\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip\n+ batch_size: 1\n+\n+MOT:\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip\n+ tracker_config: deploy/pipeline/config/tracker_config.yml\n+ batch_size: 1\n+ enable: True\n+\n+VEHICLE_PLATE:\n+ det_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_det_infer.tar.gz\n+ det_limit_side_len: 736\n+ det_limit_type: \"min\"\n+ rec_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_rec_infer.tar.gz\n+ rec_image_shape: [3, 48, 320]\n+ rec_batch_num: 6\n+ word_dict_path: deploy/pipeline/ppvehicle/rec_word_dict.txt\n+ enable: True\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/config/infer_cfg_ppvehicle.yml",
"new_path": "deploy/pipeline/config/infer_cfg_ppvehicle.yml",
"diff": "@@ -3,33 +3,28 @@ visual: True\nwarmup_frame: 50\nDET:\n- model_dir: output_inference/mot_ppyoloe_l_36e_ppvehicle/\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip\nbatch_size: 1\nMOT:\n- model_dir: output_inference/mot_ppyoloe_l_36e_ppvehicle/\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip\ntracker_config: deploy/pipeline/config/tracker_config.yml\nbatch_size: 1\nenable: False\nVEHICLE_PLATE:\n- det_model_dir: output_inference/ch_PP-OCRv3_det_infer/\n+ det_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_det_infer.tar.gz\ndet_limit_side_len: 736\ndet_limit_type: \"min\"\n- rec_model_dir: output_inference/ch_PP-OCRv3_rec_infer/\n+ rec_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_rec_infer.tar.gz\nrec_image_shape: [3, 48, 320]\nrec_batch_num: 6\nword_dict_path: deploy/pipeline/ppvehicle/rec_word_dict.txt\nenable: False\nVEHICLE_ATTR:\n- model_dir: output_inference/vehicle_attribute_infer/\n+ model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/vehicle_attribute_model.zip\nbatch_size: 8\ncolor_threshold: 0.5\ntype_threshold: 0.5\nenable: False\n-\n-REID:\n- model_dir: output_inference/vehicle_reid_model/\n- batch_size: 16\n- enable: False\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
use link for vehicle model (#6645)
|
499,348 |
17.08.2022 00:04:47
| -28,800 |
28199de73e51ff08f54d941c83ee24056ffd16dd
|
add per trackid time info
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipe_utils.py",
"new_path": "deploy/pipeline/pipe_utils.py",
"diff": "@@ -68,6 +68,7 @@ class PipeTimer(Times):\n'vehicleplate': Times()\n}\nself.img_num = 0\n+ self.track_num = 0\ndef get_total_time(self):\ntotal_time = self.total_time.value()\n@@ -86,8 +87,11 @@ class PipeTimer(Times):\nfor k, v in self.module_time.items():\nv_time = round(v.value(), 4)\n- if v_time > 0:\n+ if v_time > 0 and k in ['det', 'mot', 'video_action']:\nprint(\"{} time(ms): {}\".format(k, v_time * 1000))\n+ elif v_time > 0:\n+ print(\"{} time(ms): {}; per trackid average time(ms): {}\".\n+ format(k, v_time * 1000, v_time * 1000 / self.track_num))\nprint(\"average latency time(ms): {:.2f}, QPS: {:2f}\".format(\naverage_latency * 1000, qps))\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipeline.py",
"new_path": "deploy/pipeline/pipeline.py",
"diff": "@@ -598,10 +598,11 @@ class PipePredictor(object):\nif not ret:\nbreak\nframe_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)\n+ if frame_id > self.warmup_frame:\n+ self.pipe_timer.total_time.start()\nif self.modebase[\"idbased\"] or self.modebase[\"skeletonbased\"]:\nif frame_id > self.warmup_frame:\n- self.pipe_timer.total_time.start()\nself.pipe_timer.module_time['mot'].start()\nmot_skip_frame_num = self.mot_predictor.skip_frame_num\n@@ -612,11 +613,12 @@ class PipePredictor(object):\n[copy.deepcopy(frame_rgb)],\nvisual=False,\nreuse_det_result=reuse_det_result)\n- if frame_id > self.warmup_frame:\n- self.pipe_timer.module_time['mot'].end()\n# mot output format: id, class, score, xmin, ymin, xmax, ymax\nmot_res = parse_mot_res(res)\n+ if frame_id > self.warmup_frame:\n+ self.pipe_timer.module_time['mot'].end()\n+ self.pipe_timer.track_num += len(mot_res['boxes'])\n# flow_statistic only support single class MOT\nboxes, scores, ids = res[0] # batch size = 1 in MOT\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add per trackid time info (#6664)
|
499,301 |
18.08.2022 18:32:10
| -28,800 |
8fbdf1cb6bedd8a884d2f3d0482b4ef39a36f142
|
add ppyoloe plus cfgs
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_l_qat_dis.yaml",
"diff": "+\n+Global:\n+ reader_config: configs/ppyoloe_plus_reader.yml\n+ input_list: ['image', 'scale_factor']\n+ arch: YOLO\n+ Evaluation: True\n+ model_dir: ./ppyoloe_plus_crn_l_80e_coco\n+ model_filename: model.pdmodel\n+ params_filename: model.pdiparams\n+\n+Distillation:\n+ alpha: 1.0\n+ loss: soft_label\n+\n+Quantization:\n+ use_pact: true\n+ activation_quantize_type: 'moving_average_abs_max'\n+ quantize_op_types:\n+ - conv2d\n+ - depthwise_conv2d\n+\n+TrainConfig:\n+ train_iter: 5000\n+ eval_iter: 1000\n+ learning_rate:\n+ type: CosineAnnealingDecay\n+ learning_rate: 0.00003\n+ T_max: 6000\n+ optimizer_builder:\n+ optimizer:\n+ type: SGD\n+ weight_decay: 4.0e-05\n+\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_m_qat_dis.yaml",
"diff": "+\n+Global:\n+ reader_config: configs/ppyoloe_plus_reader.yml\n+ input_list: ['image', 'scale_factor']\n+ arch: YOLO\n+ Evaluation: True\n+ model_dir: ./ppyoloe_plus_crn_m_80e_coco\n+ model_filename: model.pdmodel\n+ params_filename: model.pdiparams\n+\n+Distillation:\n+ alpha: 1.0\n+ loss: soft_label\n+\n+Quantization:\n+ use_pact: true\n+ activation_quantize_type: 'moving_average_abs_max'\n+ quantize_op_types:\n+ - conv2d\n+ - depthwise_conv2d\n+\n+TrainConfig:\n+ train_iter: 5000\n+ eval_iter: 1000\n+ learning_rate:\n+ type: CosineAnnealingDecay\n+ learning_rate: 0.00003\n+ T_max: 6000\n+ optimizer_builder:\n+ optimizer:\n+ type: SGD\n+ weight_decay: 4.0e-05\n+\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_reader.yml",
"diff": "+\n+\n+metric: COCO\n+num_classes: 80\n+\n+# Datset configuration\n+TrainDataset:\n+ !COCODataSet\n+ image_dir: train2017\n+ anno_path: annotations/instances_train2017.json\n+ dataset_dir: dataset/coco/\n+\n+EvalDataset:\n+ !COCODataSet\n+ image_dir: val2017\n+ anno_path: annotations/instances_val2017.json\n+ dataset_dir: dataset/coco/\n+\n+worker_num: 0\n+\n+# preprocess reader in test\n+EvalReader:\n+ sample_transforms:\n+ - Decode: {}\n+ - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}\n+ - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], is_scale: True}\n+ - Permute: {}\n+ batch_size: 4\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_s_qat_dis.yaml",
"diff": "+\n+Global:\n+ reader_config: configs/ppyoloe_plus_reader.yml\n+ input_list: ['image', 'scale_factor']\n+ arch: YOLO\n+ Evaluation: True\n+ model_dir: ./ppyoloe_plus_crn_s_80e_coco\n+ model_filename: model.pdmodel\n+ params_filename: model.pdiparams\n+\n+Distillation:\n+ alpha: 1.0\n+ loss: soft_label\n+\n+Quantization:\n+ use_pact: true\n+ activation_quantize_type: 'moving_average_abs_max'\n+ quantize_op_types:\n+ - conv2d\n+ - depthwise_conv2d\n+\n+TrainConfig:\n+ train_iter: 5000\n+ eval_iter: 1000\n+ learning_rate:\n+ type: CosineAnnealingDecay\n+ learning_rate: 0.00003\n+ T_max: 6000\n+ optimizer_builder:\n+ optimizer:\n+ type: SGD\n+ weight_decay: 4.0e-05\n+\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_x_qat_dis.yaml",
"diff": "+\n+Global:\n+ reader_config: configs/ppyoloe_plus_reader.yml\n+ input_list: ['image', 'scale_factor']\n+ arch: YOLO\n+ Evaluation: True\n+ model_dir: ./ppyoloe_plus_crn_x_80e_coco\n+ model_filename: model.pdmodel\n+ params_filename: model.pdiparams\n+\n+Distillation:\n+ alpha: 1.0\n+ loss: soft_label\n+\n+Quantization:\n+ use_pact: true\n+ activation_quantize_type: 'moving_average_abs_max'\n+ quantize_op_types:\n+ - conv2d\n+ - depthwise_conv2d\n+\n+TrainConfig:\n+ train_iter: 5000\n+ eval_iter: 1000\n+ learning_rate:\n+ type: CosineAnnealingDecay\n+ learning_rate: 0.00003\n+ T_max: 6000\n+ optimizer_builder:\n+ optimizer:\n+ type: SGD\n+ weight_decay: 4.0e-05\n+\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add ppyoloe plus cfgs (#6686)
|
499,339 |
22.08.2022 14:36:01
| -28,800 |
10e7fe232c83dacee0f517d78644b705e5d24a57
|
[deploy] alter save coco format json in deploy/python/infer.py
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -36,7 +36,7 @@ from picodet_postprocess import PicoDetPostProcess\nfrom preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, WarpAffine, Pad, decode_image\nfrom keypoint_preprocess import EvalAffine, TopDownEvalAffine, expand_crop\nfrom visualize import visualize_box_mask\n-from utils import argsparser, Timer, get_current_memory_mb, multiclass_nms\n+from utils import argsparser, Timer, get_current_memory_mb, multiclass_nms, coco_clsid2catid\n# Global dictionary\nSUPPORT_MODELS = {\n@@ -226,7 +226,7 @@ class Detector(object):\nmatch_threshold=0.6,\nmatch_metric='iou',\nvisual=True,\n- save_file=None):\n+ save_results=False):\n# slice infer only support bs=1\nresults = []\ntry:\n@@ -295,14 +295,13 @@ class Detector(object):\nthreshold=self.threshold)\nresults.append(merged_results)\n- if visual:\nprint('Test iter {}'.format(i))\n- if save_file is not None:\n- Path(self.output_dir).mkdir(exist_ok=True)\n- self.format_coco_results(image_list, results, save_file=save_file)\n-\nresults = self.merge_batch_result(results)\n+ if save_results:\n+ Path(self.output_dir).mkdir(exist_ok=True)\n+ self.save_coco_results(\n+ img_list, results, use_coco_category=FLAGS.use_coco_category)\nreturn results\ndef predict_image(self,\n@@ -310,7 +309,7 @@ class Detector(object):\nrun_benchmark=False,\nrepeats=1,\nvisual=True,\n- save_file=None):\n+ save_results=False):\nbatch_loop_cnt = math.ceil(float(len(image_list)) / self.batch_size)\nresults = []\nfor i in range(batch_loop_cnt):\n@@ -367,14 +366,13 @@ class Detector(object):\nthreshold=self.threshold)\nresults.append(result)\n- if visual:\nprint('Test iter {}'.format(i))\n- if save_file is not None:\n- Path(self.output_dir).mkdir(exist_ok=True)\n- self.format_coco_results(image_list, results, save_file=save_file)\n-\nresults = self.merge_batch_result(results)\n+ if save_results:\n+ Path(self.output_dir).mkdir(exist_ok=True)\n+ self.save_coco_results(\n+ image_list, results, use_coco_category=FLAGS.use_coco_category)\nreturn results\ndef predict_video(self, video_file, camera_id):\n@@ -418,67 +416,62 @@ class Detector(object):\nbreak\nwriter.release()\n- @staticmethod\n- def format_coco_results(image_list, results, save_file=None):\n- coco_results = []\n- image_id = 0\n-\n- for result in results:\n- start_idx = 0\n- for box_num in result['boxes_num']:\n- idx_slice = slice(start_idx, start_idx + box_num)\n- start_idx += box_num\n-\n- image_file = image_list[image_id]\n- image_id += 1\n-\n- if 'boxes' in result:\n- boxes = result['boxes'][idx_slice, :]\n- per_result = [\n- {\n- 'image_file': image_file,\n- 'bbox':\n- [box[2], box[3], box[4] - box[2],\n+ def save_coco_results(self, image_list, results, use_coco_category=False):\n+ bbox_results = []\n+ mask_results = []\n+ idx = 0\n+ print(\"Start saving coco json files...\")\n+ for i, box_num in enumerate(results['boxes_num']):\n+ file_name = os.path.split(image_list[i])[-1]\n+ if use_coco_category:\n+ img_id = int(os.path.splitext(file_name)[0])\n+ else:\n+ img_id = i\n+\n+ if 'boxes' in results:\n+ boxes = results['boxes'][idx:idx + box_num].tolist()\n+ bbox_results.extend([{\n+ 'image_id': img_id,\n+ 'category_id': coco_clsid2catid[int(box[0])] \\\n+ if use_coco_category else int(box[0]),\n+ 'file_name': file_name,\n+ 'bbox': [box[2], box[3], box[4] - box[2],\nbox[5] - box[3]], # xyxy -> xywh\n- 'score': box[1],\n- 'category_id': int(box[0]),\n- } for k, box in enumerate(boxes.tolist())\n- ]\n+ 'score': box[1]} for box in boxes])\n- elif 'segm' in result:\n+ if 'masks' in results:\nimport pycocotools.mask as mask_util\n- scores = result['score'][idx_slice].tolist()\n- category_ids = result['label'][idx_slice].tolist()\n- segms = result['segm'][idx_slice, :]\n- rles = [\n- mask_util.encode(\n+ boxes = results['boxes'][idx:idx + box_num].tolist()\n+ masks = results['masks'][i][:box_num].astype(np.uint8)\n+ seg_res = []\n+ for box, mask in zip(boxes, masks):\n+ rle = mask_util.encode(\nnp.array(\n- mask[:, :, np.newaxis],\n- dtype=np.uint8,\n- order='F'))[0] for mask in segms\n- ]\n- for rle in rles:\n- rle['counts'] = rle['counts'].decode('utf-8')\n-\n- per_result = [{\n- 'image_file': image_file,\n+ mask[:, :, None], dtype=np.uint8, order=\"F\"))[0]\n+ if 'counts' in rle:\n+ rle['counts'] = rle['counts'].decode(\"utf8\")\n+ seg_res.append({\n+ 'image_id': img_id,\n+ 'category_id': coco_clsid2catid[int(box[0])] \\\n+ if use_coco_category else int(box[0]),\n+ 'file_name': file_name,\n'segmentation': rle,\n- 'score': scores[k],\n- 'category_id': category_ids[k],\n- } for k, rle in enumerate(rles)]\n+ 'score': box[1]})\n+ mask_results.extend(seg_res)\n- else:\n- raise RuntimeError('')\n-\n- # per_result = [item for item in per_result if item['score'] > threshold]\n- coco_results.extend(per_result)\n+ idx += box_num\n- if save_file:\n- with open(os.path.join(save_file), 'w') as f:\n- json.dump(coco_results, f)\n-\n- return coco_results\n+ if bbox_results:\n+ bbox_file = os.path.join(self.output_dir, \"bbox.json\")\n+ with open(bbox_file, 'w') as f:\n+ json.dump(bbox_results, f)\n+ print(f\"The bbox result is saved to {bbox_file}\")\n+ if mask_results:\n+ mask_file = os.path.join(self.output_dir, \"mask.json\")\n+ with open(mask_file, 'w') as f:\n+ json.dump(mask_results, f)\n+ print(f\"The mask result is saved to {mask_file}\")\nclass DetectorSOLOv2(Detector):\n@@ -956,8 +949,6 @@ def main():\nif FLAGS.image_dir is None and FLAGS.image_file is not None:\nassert FLAGS.batch_size == 1, \"batch_size should be 1, when image_file is not None\"\nimg_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)\n- save_file = os.path.join(FLAGS.output_dir,\n- 'results.json') if FLAGS.save_results else None\nif FLAGS.slice_infer:\ndetector.predict_image_slice(\nimg_list,\n@@ -966,10 +957,15 @@ def main():\nFLAGS.combine_method,\nFLAGS.match_threshold,\nFLAGS.match_metric,\n- save_file=save_file)\n+ visual=FLAGS.save_images,\n+ save_results=FLAGS.save_results)\nelse:\ndetector.predict_image(\n- img_list, FLAGS.run_benchmark, repeats=100, save_file=save_file)\n+ img_list,\n+ FLAGS.run_benchmark,\n+ repeats=100,\n+ visual=FLAGS.save_images,\n+ save_results=FLAGS.save_results)\nif not FLAGS.run_benchmark:\ndetector.det_times.info(average=True)\nelse:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/utils.py",
"new_path": "deploy/python/utils.py",
"diff": "@@ -109,6 +109,7 @@ def argsparser():\nparser.add_argument(\n'--save_images',\naction='store_true',\n+ default=False,\nhelp='Save visualization image results.')\nparser.add_argument(\n'--save_mot_txts',\n@@ -159,9 +160,14 @@ def argsparser():\nhelp=\"Whether do random padding for action recognition.\")\nparser.add_argument(\n\"--save_results\",\n- type=bool,\n+ action='store_true',\ndefault=False,\nhelp=\"Whether save detection result to file using coco format\")\n+ parser.add_argument(\n+ '--use_coco_category',\n+ action='store_true',\n+ default=False,\n+ help='Whether to use the coco format dictionary `clsid2catid`')\nparser.add_argument(\n\"--slice_infer\",\naction='store_true',\n@@ -386,3 +392,87 @@ def nms(dets, match_threshold=0.6, match_metric='iou'):\nkeep = np.where(suppressed == 0)[0]\ndets = dets[keep, :]\nreturn dets\n+\n+\n+coco_clsid2catid = {\n+ 0: 1,\n+ 1: 2,\n+ 2: 3,\n+ 3: 4,\n+ 4: 5,\n+ 5: 6,\n+ 6: 7,\n+ 7: 8,\n+ 8: 9,\n+ 9: 10,\n+ 10: 11,\n+ 11: 13,\n+ 12: 14,\n+ 13: 15,\n+ 14: 16,\n+ 15: 17,\n+ 16: 18,\n+ 17: 19,\n+ 18: 20,\n+ 19: 21,\n+ 20: 22,\n+ 21: 23,\n+ 22: 24,\n+ 23: 25,\n+ 24: 27,\n+ 25: 28,\n+ 26: 31,\n+ 27: 32,\n+ 28: 33,\n+ 29: 34,\n+ 30: 35,\n+ 31: 36,\n+ 32: 37,\n+ 33: 38,\n+ 34: 39,\n+ 35: 40,\n+ 36: 41,\n+ 37: 42,\n+ 38: 43,\n+ 39: 44,\n+ 40: 46,\n+ 41: 47,\n+ 42: 48,\n+ 43: 49,\n+ 44: 50,\n+ 45: 51,\n+ 46: 52,\n+ 47: 53,\n+ 48: 54,\n+ 49: 55,\n+ 50: 56,\n+ 51: 57,\n+ 52: 58,\n+ 53: 59,\n+ 54: 60,\n+ 55: 61,\n+ 56: 62,\n+ 57: 63,\n+ 58: 64,\n+ 59: 65,\n+ 60: 67,\n+ 61: 70,\n+ 62: 72,\n+ 63: 73,\n+ 64: 74,\n+ 65: 75,\n+ 66: 76,\n+ 67: 77,\n+ 68: 78,\n+ 69: 79,\n+ 70: 80,\n+ 71: 81,\n+ 72: 82,\n+ 73: 84,\n+ 74: 85,\n+ 75: 86,\n+ 76: 87,\n+ 77: 88,\n+ 78: 89,\n+ 79: 90\n+}\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[deploy] alter save coco format json in deploy/python/infer.py (#6705)
|
499,339 |
22.08.2022 17:28:45
| -28,800 |
b84d8fd65dfc8ee608bc6bde5bdee1cff4d11d4f
|
[ppyoloe-plus] update ppyoloe legacy configs
|
[
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/README.md",
"new_path": "configs/ppyoloe/README_legacy.md",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/_base_/optimizer_300e.yml",
"new_path": "configs/ppyoloe/_base_/optimizer_300e.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/_base_/optimizer_36e_xpu.yml",
"new_path": "configs/ppyoloe/_base_/optimizer_36e_xpu.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/_base_/optimizer_400e.yml",
"new_path": "configs/ppyoloe/_base_/optimizer_400e.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/_base_/ppyoloe_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/_base_/ppyoloe_reader.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_reader.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/ppyoloe_crn_l_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml",
"diff": "_BASE_: [\n- '../../datasets/coco_detection.yml',\n- '../../runtime.yml',\n+ '../datasets/coco_detection.yml',\n+ '../runtime.yml',\n'./_base_/optimizer_300e.yml',\n'./_base_/ppyoloe_crn.yml',\n'./_base_/ppyoloe_reader.yml',\n"
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/ppyoloe_crn_l_36e_coco_xpu.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_l_36e_coco_xpu.yml",
"diff": "_BASE_: [\n- '../../datasets/coco_detection.yml',\n- '../../runtime.yml',\n+ '../datasets/coco_detection.yml',\n+ '../runtime.yml',\n'./_base_/optimizer_36e_xpu.yml',\n'./_base_/ppyoloe_reader.yml',\n]\n"
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/ppyoloe_crn_m_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_m_300e_coco.yml",
"diff": "_BASE_: [\n- '../../datasets/coco_detection.yml',\n- '../../runtime.yml',\n+ '../datasets/coco_detection.yml',\n+ '../runtime.yml',\n'./_base_/optimizer_300e.yml',\n'./_base_/ppyoloe_crn.yml',\n'./_base_/ppyoloe_reader.yml',\n"
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/ppyoloe_crn_s_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml",
"diff": "_BASE_: [\n- '../../datasets/coco_detection.yml',\n- '../../runtime.yml',\n+ '../datasets/coco_detection.yml',\n+ '../runtime.yml',\n'./_base_/optimizer_300e.yml',\n'./_base_/ppyoloe_crn.yml',\n'./_base_/ppyoloe_reader.yml',\n"
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/ppyoloe_crn_s_400e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_s_400e_coco.yml",
"diff": "_BASE_: [\n- '../../datasets/coco_detection.yml',\n- '../../runtime.yml',\n+ '../datasets/coco_detection.yml',\n+ '../runtime.yml',\n'./_base_/optimizer_400e.yml',\n'./_base_/ppyoloe_crn.yml',\n'./_base_/ppyoloe_reader.yml',\n"
},
{
"change_type": "RENAME",
"old_path": "configs/ppyoloe/legacy_model/ppyoloe_crn_x_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_x_300e_coco.yml",
"diff": "_BASE_: [\n- '../../datasets/coco_detection.yml',\n- '../../runtime.yml',\n+ '../datasets/coco_detection.yml',\n+ '../runtime.yml',\n'./_base_/optimizer_300e.yml',\n'./_base_/ppyoloe_crn.yml',\n'./_base_/ppyoloe_reader.yml',\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[ppyoloe-plus] update ppyoloe legacy configs (#6718)
|
499,298 |
23.08.2022 22:27:32
| -28,800 |
4708b0811ee99bf9d4cf23f309817675191a434a
|
fix iters less than batchsize in warmup
|
[
{
"change_type": "MODIFY",
"old_path": "configs/mot/fairmot/_base_/optimizer_30e_momentum.yml",
"new_path": "configs/mot/fairmot/_base_/optimizer_30e_momentum.yml",
"diff": "@@ -7,8 +7,9 @@ LearningRate:\ngamma: 0.1\nmilestones: [15, 22]\nuse_warmup: True\n- - !BurninWarmup\n+ - !ExpWarmup\nsteps: 1000\n+ power: 4\nOptimizerBuilder:\noptimizer:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/jde/_base_/optimizer_30e.yml",
"new_path": "configs/mot/jde/_base_/optimizer_30e.yml",
"diff": "@@ -7,8 +7,9 @@ LearningRate:\ngamma: 0.1\nmilestones: [15, 22]\nuse_warmup: True\n- - !BurninWarmup\n+ - !ExpWarmup\nsteps: 1000\n+ power: 4\nOptimizerBuilder:\noptimizer:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/jde/_base_/optimizer_60e.yml",
"new_path": "configs/mot/jde/_base_/optimizer_60e.yml",
"diff": "@@ -7,8 +7,9 @@ LearningRate:\ngamma: 0.1\nmilestones: [30, 44]\nuse_warmup: True\n- - !BurninWarmup\n+ - !ExpWarmup\nsteps: 1000\n+ power: 4\nOptimizerBuilder:\noptimizer:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/mcfairmot/mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_vehicle_bytetracker.yml",
"new_path": "configs/mot/mcfairmot/mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_vehicle_bytetracker.yml",
"diff": "@@ -63,8 +63,9 @@ LearningRate:\ngamma: 0.1\nmilestones: [15, 22]\nuse_warmup: True\n- - !BurninWarmup\n+ - !ExpWarmup\nsteps: 1000\n+ power: 4\nOptimizerBuilder:\noptimizer:\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -150,6 +150,10 @@ class Trainer(object):\n# build optimizer in train mode\nif self.mode == 'train':\nsteps_per_epoch = len(self.loader)\n+ if steps_per_epoch < 1:\n+ logger.warning(\n+ \"Samples in dataset are less than batch_size, please set smaller batch_size in TrainReader.\"\n+ )\nself.lr = create('LearningRate')(steps_per_epoch)\nself.optimizer = create('OptimizerBuilder')(self.lr, self.model)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer/optimizer.py",
"new_path": "ppdet/optimizer/optimizer.py",
"diff": "@@ -176,6 +176,7 @@ class LinearWarmup(object):\nvalue = []\nwarmup_steps = self.epochs * step_per_epoch \\\nif self.epochs is not None else self.steps\n+ warmup_steps = max(warmup_steps, 1)\nfor i in range(warmup_steps + 1):\nif warmup_steps > 0:\nalpha = i / warmup_steps\n@@ -187,31 +188,6 @@ class LinearWarmup(object):\nreturn boundary, value\n-@serializable\n-class BurninWarmup(object):\n- \"\"\"\n- Warm up learning rate in burnin mode\n- Args:\n- steps (int): warm up steps\n- \"\"\"\n-\n- def __init__(self, steps=1000):\n- super(BurninWarmup, self).__init__()\n- self.steps = steps\n-\n- def __call__(self, base_lr, step_per_epoch):\n- boundary = []\n- value = []\n- burnin = min(self.steps, step_per_epoch)\n- for i in range(burnin + 1):\n- factor = (i * 1.0 / burnin)**4\n- lr = base_lr * factor\n- value.append(lr)\n- if i > 0:\n- boundary.append(i)\n- return boundary, value\n-\n-\n@serializable\nclass ExpWarmup(object):\n\"\"\"\n@@ -220,19 +196,22 @@ class ExpWarmup(object):\nsteps (int): warm up steps.\nepochs (int|None): use epochs as warm up steps, the priority\nof `epochs` is higher than `steps`. Default: None.\n+ power (int): Exponential coefficient. Default: 2.\n\"\"\"\n- def __init__(self, steps=5, epochs=None):\n+ def __init__(self, steps=1000, epochs=None, power=2):\nsuper(ExpWarmup, self).__init__()\nself.steps = steps\nself.epochs = epochs\n+ self.power = power\ndef __call__(self, base_lr, step_per_epoch):\nboundary = []\nvalue = []\nwarmup_steps = self.epochs * step_per_epoch if self.epochs is not None else self.steps\n+ warmup_steps = max(warmup_steps, 1)\nfor i in range(warmup_steps + 1):\n- factor = (i / float(warmup_steps))**2\n+ factor = (i / float(warmup_steps))**self.power\nvalue.append(base_lr * factor)\nif i > 0:\nboundary.append(i)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix iters less than batchsize in warmup (#6724)
|
499,364 |
24.08.2022 10:17:08
| -28,800 |
da3a1fc995d5e0f35a0d86c222ca9845f62c9d82
|
fix distill bug
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/slim/distill.py",
"new_path": "ppdet/slim/distill.py",
"diff": "@@ -436,7 +436,7 @@ class FGDFeatureLoss(nn.Layer):\nMask_bg = paddle.ones_like(tea_spatial_att)\none_tmp = paddle.ones([*tea_spatial_att.shape[1:]])\nzero_tmp = paddle.zeros([*tea_spatial_att.shape[1:]])\n- mask_fg.stop_gradient = True\n+ Mask_fg.stop_gradient = True\nMask_bg.stop_gradient = True\none_tmp.stop_gradient = True\nzero_tmp.stop_gradient = True\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix distill bug (#6733)
|
499,339 |
24.08.2022 12:15:33
| -28,800 |
8c790b977e241ad595dc2864a2d12ae10d7efbd3
|
[docs] update ppyoloe_plus docs, test=document_fix
|
[
{
"change_type": "DELETE",
"old_path": "docs/images/ppyoloe_map_fps.png",
"new_path": "docs/images/ppyoloe_map_fps.png",
"diff": "Binary files a/docs/images/ppyoloe_map_fps.png and /dev/null differ\n"
},
{
"change_type": "ADD",
"old_path": "docs/images/ppyoloe_plus_map_fps.png",
"new_path": "docs/images/ppyoloe_plus_map_fps.png",
"diff": "Binary files /dev/null and b/docs/images/ppyoloe_plus_map_fps.png differ\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[docs] update ppyoloe_plus docs, test=document_fix (#6729)
|
499,298 |
29.08.2022 14:05:46
| -28,800 |
73ef70f030955d2bfb88a0a3844668228643eba2
|
[cherry-pick] fix illegal parking doc
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pipeline.py",
"new_path": "deploy/pipeline/pipeline.py",
"diff": "@@ -626,10 +626,20 @@ class PipePredictor(object):\nmot_result = (frame_id + 1, boxes[0], scores[0],\nids[0]) # single class\nstatistic = flow_statistic(\n- mot_result, self.secs_interval, self.do_entrance_counting,\n- self.do_break_in_counting, self.region_type, video_fps,\n- entrance, id_set, interval_id_set, in_id_list, out_id_list,\n- prev_center, records)\n+ mot_result,\n+ self.secs_interval,\n+ self.do_entrance_counting,\n+ self.do_break_in_counting,\n+ self.region_type,\n+ video_fps,\n+ entrance,\n+ id_set,\n+ interval_id_set,\n+ in_id_list,\n+ out_id_list,\n+ prev_center,\n+ records,\n+ ids2names=self.mot_predictor.pred_config.labels)\nrecords = statistic['records']\nif self.illegal_parking_time != -1:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot/utils.py",
"new_path": "deploy/pptracking/python/mot/utils.py",
"diff": "@@ -224,7 +224,7 @@ def flow_statistic(result,\nprev_center,\nrecords,\ndata_type='mot',\n- num_classes=1):\n+ ids2names=['pedestrian']):\n# Count in/out number:\n# Note that 'region_type' should be one of ['horizontal', 'vertical', 'custom'],\n# 'horizontal' and 'vertical' means entrance is the center line as the entrance when do_entrance_counting,\n@@ -282,25 +282,27 @@ def flow_statistic(result,\nframe_id -= 1\nx1, y1, w, h = tlwh\ncenter_x = min(x1 + w / 2., im_w - 1)\n- center_down_y = min(y1 + h, im_h - 1)\n+ if ids2names[0] == 'pedestrian':\n+ center_y = min(y1 + h, im_h - 1)\n+ else:\n+ center_y = min(y1 + h / 2, im_h - 1)\n# counting objects in region of the first frame\nif frame_id == 1:\n- if in_quadrangle([center_x, center_down_y], entrance, im_h,\n- im_w):\n+ if in_quadrangle([center_x, center_y], entrance, im_h, im_w):\nin_id_list.append(-1)\nelse:\n- prev_center[track_id] = [center_x, center_down_y]\n+ prev_center[track_id] = [center_x, center_y]\nelse:\nif track_id in prev_center:\nif not in_quadrangle(prev_center[track_id], entrance, im_h,\nim_w) and in_quadrangle(\n- [center_x, center_down_y],\n- entrance, im_h, im_w):\n+ [center_x, center_y], entrance,\n+ im_h, im_w):\nin_id_list.append(track_id)\n- prev_center[track_id] = [center_x, center_down_y]\n+ prev_center[track_id] = [center_x, center_y]\nelse:\n- prev_center[track_id] = [center_x, center_down_y]\n+ prev_center[track_id] = [center_x, center_y]\n# Count totol number, number at a manual-setting interval\nframe_id, tlwhs, tscores, track_ids = result\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot_jde_infer.py",
"new_path": "deploy/pptracking/python/mot_jde_infer.py",
"diff": "@@ -393,10 +393,21 @@ class JDE_Detector(Detector):\nresult = (frame_id + 1, online_tlwhs[0], online_scores[0],\nonline_ids[0])\nstatistic = flow_statistic(\n- result, self.secs_interval, self.do_entrance_counting,\n- self.do_break_in_counting, self.region_type, video_fps,\n- entrance, id_set, interval_id_set, in_id_list, out_id_list,\n- prev_center, records, data_type, num_classes)\n+ result,\n+ self.secs_interval,\n+ self.do_entrance_counting,\n+ self.do_break_in_counting,\n+ self.region_type,\n+ video_fps,\n+ entrance,\n+ id_set,\n+ interval_id_set,\n+ in_id_list,\n+ out_id_list,\n+ prev_center,\n+ records,\n+ data_type,\n+ ids2names=self.pred_config.labels)\nrecords = statistic['records']\nfps = 1. / timer.duration\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot_sde_infer.py",
"new_path": "deploy/pptracking/python/mot_sde_infer.py",
"diff": "@@ -634,10 +634,21 @@ class SDE_Detector(Detector):\nresult = (frame_id + 1, online_tlwhs[0], online_scores[0],\nonline_ids[0])\nstatistic = flow_statistic(\n- result, self.secs_interval, self.do_entrance_counting,\n- self.do_break_in_counting, self.region_type, video_fps,\n- entrance, id_set, interval_id_set, in_id_list, out_id_list,\n- prev_center, records, data_type, num_classes)\n+ result,\n+ self.secs_interval,\n+ self.do_entrance_counting,\n+ self.do_break_in_counting,\n+ self.region_type,\n+ video_fps,\n+ entrance,\n+ id_set,\n+ interval_id_set,\n+ in_id_list,\n+ out_id_list,\n+ prev_center,\n+ records,\n+ data_type,\n+ ids2names=self.pred_config.labels)\nrecords = statistic['records']\nfps = 1. / timer.duration\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] fix illegal parking doc (#6764)
|
499,339 |
29.08.2022 14:18:33
| -28,800 |
f52c63bf9f4c5dcbc9352e3f1bcf527ffd1c61ef
|
[cherry-pick] fix params save_images
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/utils.py",
"new_path": "deploy/python/utils.py",
"diff": "@@ -108,8 +108,8 @@ def argsparser():\n\"calibration, trt_calib_mode need to set True.\")\nparser.add_argument(\n'--save_images',\n- action='store_true',\n- default=False,\n+ type=bool,\n+ default=True,\nhelp='Save visualization image results.')\nparser.add_argument(\n'--save_mot_txts',\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] fix params save_images (#6779)
|
499,299 |
29.08.2022 15:36:48
| -28,800 |
2382374e7d2161aa95a33c28be4ad1c75b17ae58
|
fix bug in pipeline doc and update kpt training config
|
[
{
"change_type": "RENAME",
"old_path": "configs/pphuman/hrnet_w32_256x192.yml",
"new_path": "configs/pphuman/dark_hrnet_w32_256x192.yml",
"diff": "@@ -101,7 +101,7 @@ TrainReader:\nflip_pairs: *flip_perm\n- TopDownAffine:\ntrainsize: *trainsize\n- - ToHeatmapsTopDown:\n+ - ToHeatmapsTopDown_DARK:\nhmsize: *hmsize\nsigma: 2\nbatch_transforms:\n@@ -125,6 +125,7 @@ EvalReader:\nis_scale: true\n- Permute: {}\nbatch_size: 16\n+ drop_empty: false\nTestReader:\ninputs_def:\n@@ -139,4 +140,3 @@ TestReader:\nis_scale: true\n- Permute: {}\nbatch_size: 1\n- fuse_normalize: false #whether to fuse nomalize layer into model while export model\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/docs/tutorials/ppvehicle_attribute_en.md",
"new_path": "deploy/pipeline/docs/tutorials/ppvehicle_attribute_en.md",
"diff": "@@ -6,12 +6,12 @@ Vehicle attribute recognition is widely used in smart cities, smart transportati\n| Task | Algorithm | Precision | Inference Speed | Download |\n|-----------|------|-----------|----------|---------------------|\n-| Vehicle Detection/Tracking | PP-YOLOE | - | - | [Inference and Deployment Model](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip) |\n-| Vehicle Attribute Recognition | PPLCNet | 90.81 | 2.36 ms | [Inference and Deployment Model](https://bj.bcebos.com/v1/paddledet/models/pipeline/vehicle_attribute_model.zip) |\n+| Vehicle Detection/Tracking | PP-YOLOE | mAP 63.9 | 38.67ms | [Inference and Deployment Model](https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_ppvehicle.zip) |\n+| Vehicle Attribute Recognition | PPLCNet | 90.81 | 7.31 ms | [Inference and Deployment Model](https://bj.bcebos.com/v1/paddledet/models/pipeline/vehicle_attribute_model.zip) |\nNote:\n-1. The inference speed of the attribute model is obtained from the test on Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz, with the MKLDNN acceleration strategy enabled, and 10 threads.\n+1. The inference speed of the attribute model is obtained from the test on NVIDIA T4, with TensorRT FP16. The time includes data pre-process, model inference and post-process.\n2. For introductions, please refer to [PP-LCNet Series](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.4/docs/en/models/PP-LCNet_en.md). Related paper is available on PP-LCNet paper\n3. The training and test phase of vehicle attribute recognition model are both obtained from [VeRi dataset](https://www.v7labs.com/open-datasets/veri-dataset).\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix bug in pipeline doc and update kpt training config (#6769) (#6783)
|
499,374 |
29.08.2022 18:01:57
| -28,800 |
b10ef7d9290562c90962032ae9df55849bbb22b2
|
[cherry-pick] fix recursive call of DLA
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/necks/centernet_fpn.py",
"new_path": "ppdet/modeling/necks/centernet_fpn.py",
"diff": "@@ -164,11 +164,11 @@ class IDAUp(nn.Layer):\nfor i in range(start_level + 1, end_level):\nupsample = getattr(self, 'up_' + str(i - start_level))\nproject = getattr(self, 'proj_' + str(i - start_level))\n-\ninputs[i] = project(inputs[i])\ninputs[i] = upsample(inputs[i])\nnode = getattr(self, 'node_' + str(i - start_level))\ninputs[i] = node(paddle.add(inputs[i], inputs[i - 1]))\n+ return inputs\nclass DLAUp(nn.Layer):\n@@ -197,8 +197,8 @@ class DLAUp(nn.Layer):\nout = [inputs[-1]] # start with 32\nfor i in range(len(inputs) - self.start_level - 1):\nida = getattr(self, 'ida_{}'.format(i))\n- ida(inputs, len(inputs) - i - 2, len(inputs))\n- out.insert(0, inputs[-1])\n+ outputs = ida(inputs, len(inputs) - i - 2, len(inputs))\n+ out.insert(0, outputs[-1])\nreturn out\n@@ -259,7 +259,9 @@ class CenterNetDLAFPN(nn.Layer):\ndef forward(self, body_feats):\n- dla_up_feats = self.dla_up(body_feats)\n+ inputs = [body_feats[i] for i in range(len(body_feats))]\n+\n+ dla_up_feats = self.dla_up(inputs)\nida_up_feats = []\nfor i in range(self.last_level - self.first_level):\n@@ -271,7 +273,11 @@ class CenterNetDLAFPN(nn.Layer):\nif self.with_sge:\nfeat = self.sge_attention(feat)\nif self.down_ratio != 4:\n- feat = F.interpolate(feat, scale_factor=self.down_ratio // 4, mode=\"bilinear\", align_corners=True)\n+ feat = F.interpolate(\n+ feat,\n+ scale_factor=self.down_ratio // 4,\n+ mode=\"bilinear\",\n+ align_corners=True)\nreturn feat\n@property\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] fix recursive call of DLA (#6771) (#6786)
|
499,339 |
30.08.2022 13:46:51
| -28,800 |
d02bcd932c2f76bc1dd52e7c7c593743a7f0e8b7
|
[cherry-pick] fix ppyoloe amp bug, add reduce_mean to custom_black_list
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/README.md",
"new_path": "configs/ppyoloe/README.md",
"diff": "@@ -78,12 +78,12 @@ The PaddleDetection team provides configs and weights of various feature detecti\nTraining PP-YOLOE+ on 8 GPUs with following command\n```bash\n-python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml\n+python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml --eval --amp\n```\n**Notes:**\n-- use `--amp` to train with default config to avoid out of memeory.\n- If you need to evaluate while training, please add `--eval`.\n+- PP-YOLOE+ supports mixed precision training, please add `--amp`.\n- PaddleDetection supports multi-machine distribued training, you can refer to [DistributedTraining tutorial](../../docs/DistributedTraining_en.md).\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"diff": "@@ -2,6 +2,7 @@ architecture: YOLOv3\nnorm_type: sync_bn\nuse_ema: true\nema_decay: 0.9998\n+custom_black_list: ['reduce_mean']\nYOLOv3:\nbackbone: CSPResNet\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_plus_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_plus_crn.yml",
"diff": "@@ -2,6 +2,7 @@ architecture: YOLOv3\nnorm_type: sync_bn\nuse_ema: true\nema_decay: 0.9998\n+custom_black_list: ['reduce_mean']\nYOLOv3:\nbackbone: CSPResNet\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] fix ppyoloe amp bug, add reduce_mean to custom_black_list (#6797)
|
499,327 |
30.08.2022 16:39:04
| -28,800 |
42c7468b5f0d59d8da5830d793a4cb07312000d0
|
[cherry-pick] support for edgeboard
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/export_utils.py",
"new_path": "ppdet/engine/export_utils.py",
"diff": "@@ -131,12 +131,15 @@ def _dump_infer_config(config, path, image_shape, model):\n'use_dynamic_shape': use_dynamic_shape\n})\nexport_onnx = config.get('export_onnx', False)\n+ export_eb = config.get('export_eb', False)\ninfer_arch = config['architecture']\nif 'RCNN' in infer_arch and export_onnx:\nlogger.warning(\n\"Exporting RCNN model to ONNX only support batch_size = 1\")\ninfer_cfg['export_onnx'] = True\n+ infer_cfg['export_eb'] = export_eb\n+\nif infer_arch in MOT_ARCH:\nif infer_arch == 'DeepSORT':\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/post_process.py",
"new_path": "ppdet/modeling/post_process.py",
"diff": "@@ -34,16 +34,17 @@ __all__ = [\n@register\nclass BBoxPostProcess(object):\n- __shared__ = ['num_classes', 'export_onnx']\n+ __shared__ = ['num_classes', 'export_onnx', 'export_eb']\n__inject__ = ['decode', 'nms']\ndef __init__(self, num_classes=80, decode=None, nms=None,\n- export_onnx=False):\n+ export_onnx=False, export_eb=False):\nsuper(BBoxPostProcess, self).__init__()\nself.num_classes = num_classes\nself.decode = decode\nself.nms = nms\nself.export_onnx = export_onnx\n+ self.export_eb = export_eb\ndef __call__(self, head_out, rois, im_shape, scale_factor):\n\"\"\"\n@@ -100,6 +101,10 @@ class BBoxPostProcess(object):\npred_result (Tensor): The final prediction results with shape [N, 6]\nincluding labels, scores and bboxes.\n\"\"\"\n+ if self.export_eb:\n+ # enable rcnn models for edgeboard hw to skip the following postprocess.\n+ return bboxes, bboxes, bbox_num\n+\nif not self.export_onnx:\nbboxes_list = []\nbbox_num_list = []\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] support for edgeboard #6719 (#6798)
|
499,348 |
30.08.2022 17:15:05
| -28,800 |
ede22043927a944bb4cbea0e9455dd9c91b295f0
|
update demo&fix mtmct vis; fix=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "demo/car.jpg",
"new_path": "demo/car.jpg",
"diff": "Binary files a/demo/car.jpg and b/demo/car.jpg differ\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/config/examples/infer_cfg_illegal_parking.yml",
"new_path": "deploy/pipeline/config/examples/infer_cfg_illegal_parking.yml",
"diff": "@@ -11,7 +11,7 @@ MOT:\nVEHICLE_PLATE:\ndet_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_det_infer.tar.gz\ndet_limit_side_len: 736\n- det_limit_type: \"max\"\n+ det_limit_type: \"min\"\nrec_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_rec_infer.tar.gz\nrec_image_shape: [3, 48, 320]\nrec_batch_num: 6\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/config/examples/infer_cfg_vehicle_plate.yml",
"new_path": "deploy/pipeline/config/examples/infer_cfg_vehicle_plate.yml",
"diff": "@@ -15,7 +15,7 @@ MOT:\nVEHICLE_PLATE:\ndet_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_det_infer.tar.gz\ndet_limit_side_len: 736\n- det_limit_type: \"max\"\n+ det_limit_type: \"min\"\nrec_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_rec_infer.tar.gz\nrec_image_shape: [3, 48, 320]\nrec_batch_num: 6\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/config/infer_cfg_ppvehicle.yml",
"new_path": "deploy/pipeline/config/infer_cfg_ppvehicle.yml",
"diff": "@@ -16,7 +16,7 @@ MOT:\nVEHICLE_PLATE:\ndet_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_det_infer.tar.gz\ndet_limit_side_len: 736\n- det_limit_type: \"max\"\n+ det_limit_type: \"min\"\nrec_model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ch_PP-OCRv3_rec_infer.tar.gz\nrec_image_shape: [3, 48, 320]\nrec_batch_num: 6\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pipeline/pphuman/mtmct.py",
"new_path": "deploy/pipeline/pphuman/mtmct.py",
"diff": "@@ -148,10 +148,7 @@ def save_mtmct_vis_results(camera_results, captures, output_dir,\n# add attr vis\nif multi_res:\n- tid_list = [\n- 'c' + str(idx) + '_' + 't' + str(int(j))\n- for j in range(1, len(ids) + 1)\n- ] # c0_t1, c0_t2...\n+ tid_list = multi_res.keys() # c0_t1, c0_t2...\nall_attr_result = [multi_res[i][\"attrs\"]\nfor i in tid_list] # all cid_tid result\nif any(\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update demo&fix mtmct vis; fix=document_fix (#6806)
|
499,363 |
30.08.2022 18:26:49
| -28,800 |
1b1f5909d013ffabb008eafa636a4d6cb689207d
|
optimize doc for illegal parking
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot/visualize.py",
"new_path": "deploy/pptracking/python/mot/visualize.py",
"diff": "@@ -267,6 +267,8 @@ def plot_tracking_dict(image,\nfor key, value in illegal_parking_dict.items():\nx1, y1, w, h = value['bbox']\nplate = value['plate']\n+ if plate is None:\n+ plate = \"\"\n# red box\ncv2.rectangle(im, (int(x1), int(y1)),\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
optimize doc for illegal parking (#6811)
|
499,333 |
31.08.2022 13:51:56
| -28,800 |
747c02a7a67a35ffba8a252c82951da11fffeae1
|
fix deadlink, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "README_en.md",
"new_path": "README_en.md",
"diff": "@@ -407,7 +407,6 @@ Please refer to [docs](deploy/pipeline/README_en.md) for details.\n- [Quick start](docs/tutorials/QUICK_STARTED_cn.md)\n- [Data preparation](docs/tutorials/data/README.md)\n- [Geting Started on PaddleDetection](docs/tutorials/GETTING_STARTED_cn.md)\n-- [Customize data training]((docs/tutorials/CustomizeDataTraining.md)\n- [FAQ]((docs/tutorials/FAQ)\n### Advanced tutorials\n@@ -446,7 +445,7 @@ Please refer to [docs](deploy/pipeline/README_en.md) for details.\n- [Object detection](docs/advanced_tutorials/customization/detection.md)\n- [Keypoint detection](docs/advanced_tutorials/customization/keypoint_detection.md)\n- [Multiple object tracking](docs/advanced_tutorials/customization/pphuman_mot.md)\n- - [Action recognition](docs/advanced_tutorials/customization/pphuman_action.md)\n+ - [Action recognition](docs/advanced_tutorials/customization/action_recognotion/)\n- [Attribute recognition](docs/advanced_tutorials/customization/pphuman_attribute.md)\n### Courses\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix deadlink, test=document_fix (#6816)
|
499,333 |
31.08.2022 17:46:16
| -28,800 |
32be7960c23d10bf797173059f0002de6d30ae44
|
fix unittest & whl
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/tests/test_mstest.py",
"new_path": "ppdet/modeling/tests/test_mstest.py",
"diff": "@@ -21,6 +21,7 @@ import unittest\nfrom ppdet.core.workspace import load_config\nfrom ppdet.engine import Trainer\n+\nclass TestMultiScaleInference(unittest.TestCase):\ndef setUp(self):\nself.set_config()\n@@ -48,12 +49,13 @@ class TestMultiScaleInference(unittest.TestCase):\ntests_img_root = os.path.join(os.path.dirname(__file__), 'imgs')\n# input images to predict\n- imgs = ['coco2017_val2017_000000000139.jpg', 'coco2017_val2017_000000000724.jpg']\n+ imgs = [\n+ 'coco2017_val2017_000000000139.jpg',\n+ 'coco2017_val2017_000000000724.jpg'\n+ ]\nimgs = [os.path.join(tests_img_root, img) for img in imgs]\n- trainer.predict(imgs,\n- draw_threshold=0.5,\n- output_dir='output',\n- save_txt=True)\n+ trainer.predict(\n+ imgs, draw_threshold=0.5, output_dir='output', save_results=False)\nif __name__ == '__main__':\n"
},
{
"change_type": "MODIFY",
"old_path": "scripts/build_wheel.sh",
"new_path": "scripts/build_wheel.sh",
"diff": "@@ -26,6 +26,7 @@ EGG_DIR=\"paddledet.egg-info\"\nCFG_DIR=\"configs\"\nTEST_DIR=\".tests\"\n+DATA_DIR=\"dataset\"\n# command line log config\nRED='\\033[0;31m'\n@@ -86,6 +87,7 @@ function unittest() {\n# make sure installed paddledet is used\nmkdir $TEST_DIR\ncp -r $CFG_DIR $TEST_DIR\n+ cp -r $DATA_DIR $TEST_DIR\ncd $TEST_DIR\nif [ $? != 0 ]; then\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix unittest & whl (#6820)
|
499,304 |
01.09.2022 17:04:44
| -28,800 |
da0177b404ca9f864651c3476a8aeb232d82df94
|
fix avh demo
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_avh/.gitignore",
"diff": "+include/inputs.h\n+include/outputs.h\n+\n+__pycache__/\n+build/\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_avh/Makefile",
"new_path": "deploy/third_engine/demo_avh/Makefile",
"diff": "@@ -81,13 +81,27 @@ ${BUILD_DIR}/libcmsis_startup.a: $(CMSIS_STARTUP_SRCS)\n$(QUIET)$(AR) -cr $(abspath $(BUILD_DIR)/libcmsis_startup.a) $(abspath $(BUILD_DIR))/libcmsis_startup/*.o\n$(QUIET)$(RANLIB) $(abspath $(BUILD_DIR)/libcmsis_startup.a)\n+CMSIS_SHA_FILE=${CMSIS_PATH}/977abe9849781a2e788b02282986480ff4e25ea6.sha\n+ifneq (\"$(wildcard $(CMSIS_SHA_FILE))\",\"\")\n+${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a:\n+ $(QUIET)mkdir -p $(@D)\n+ $(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)\n+ $(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all\n+else\n# Build CMSIS-NN\n${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a:\n$(QUIET)mkdir -p $(@D)\n$(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)\n$(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all\n+endif\n# Build demo application\n+ifneq (\"$(wildcard $(CMSIS_SHA_FILE))\",\"\")\n+$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \\\n+ ${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a ${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a\n+ $(QUIET)mkdir -p $(@D)\n+ $(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)\n+else\n$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \\\n${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a \\\n${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a \\\n@@ -102,6 +116,7 @@ $(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BU\n${BUILD_DIR}/cmsis_nn/Source/PoolingFunctions/libCMSISNNPooling.a\n$(QUIET)mkdir -p $(@D)\n$(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)\n+endif\nclean:\n$(QUIET)rm -rf $(BUILD_DIR)/codegen\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_avh/README.md",
"new_path": "deploy/third_engine/demo_avh/README.md",
"diff": "<!--- to you under the Apache License, Version 2.0 (the -->\n<!--- \"License\"); you may not use this file except in compliance -->\n<!--- with the License. You may obtain a copy of the License at -->\n+\n<!--- http://www.apache.org/licenses/LICENSE-2.0 -->\n+\n<!--- Unless required by applicable law or agreed to in writing, -->\n<!--- software distributed under the License is distributed on an -->\n<!--- \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->\n<!--- KIND, either express or implied. See the License for the -->\n<!--- specific language governing permissions and limitations -->\n<!--- under the License. -->\n-Running PP-PicoDet via TVM on bare metal Arm(R) Cortex(R)-M55 CPU and CMSIS-NN\n-===============================================================\n-This folder contains an example of how to use TVM to run a PP-PicoDet model\n-on bare metal Cortex(R)-M55 CPU and CMSIS-NN.\n+Running PP-PicoDet object detection model on bare metal Arm(R) Cortex(R)-M55 CPU using Arm Virtual Hardware\n+======================================================================\n+\n+This folder contains an example of how to run a PP-PicoDet model on bare metal [Cortex(R)-M55 CPU](https://www.arm.com/products/silicon-ip-cpu/cortex-m/cortex-m55) using [Arm Virtual Hardware](https://www.arm.com/products/development-tools/simulation/virtual-hardware).\n+\n-Prerequisites\n+Running environment and prerequisites\n-------------\n-If the demo is run in the ci_cpu Docker container provided with TVM, then the following\n-software will already be installed.\n+Case 1: If the demo is run in Arm Virtual Hardware Amazon Machine Image(AMI) instance hosted by [AWS](https://aws.amazon.com/marketplace/pp/prodview-urbpq7yo5va7g?sr=0-1&ref_=beagle&applicationId=AWSMPContessa)/[AWS China](https://awsmarketplace.amazonaws.cn/marketplace/pp/prodview-2y7nefntbmybu), the following software will be installed through [configure_avh.sh](./configure_avh.sh) script. It will install automatically when you run the application through [run_demo.sh](./run_demo.sh) script.\n+You can refer to this [guide](https://arm-software.github.io/AVH/main/examples/html/MicroSpeech.html#amilaunch) to launch an Arm Virtual Hardware AMI instance.\n-If the demo is not run in the ci_cpu Docker container, then you will need the following:\n+Case 2: If the demo is run in the [ci_cpu Docker container](https://github.com/apache/tvm/blob/main/docker/Dockerfile.ci_cpu) provided with [TVM](https://github.com/apache/tvm), then the following software will already be installed.\n+\n+Case 3: If the demo is not run in the ci_cpu Docker container, then you will need the following:\n- Software required to build and run the demo (These can all be installed by running\ntvm/docker/install/ubuntu_install_ethosu_driver_stack.sh.)\n- [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://release/2.5er.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)\n@@ -37,26 +42,37 @@ If the demo is not run in the ci_cpu Docker container, then you will need the fo\npip install -r ./requirements.txt\n```\n+In case2 and case3:\n+\n+You will need to update your PATH environment variable to include the path to cmake 3.19.5 and the FVP.\n+For example if you've installed these in ```/opt/arm``` , then you would do the following:\n+```bash\n+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH\n+```\n+\nYou will also need TVM which can either be:\n+ - Installed from TLCPack(see [TLCPack](https://tlcpack.ai/))\n- Built from source (see [Install from Source](https://tvm.apache.org/docs/install/from_source.html))\n- When building from source, the following need to be set in config.cmake:\n- set(USE_CMSISNN ON)\n- set(USE_MICRO ON)\n- set(USE_LLVM ON)\n- - Installed from TLCPack(see [TLCPack](https://tlcpack.ai/))\n-You will need to update your PATH environment variable to include the path to cmake 3.19.5 and the FVP.\n-For example if you've installed these in ```/opt/arm``` , then you would do the following:\n-```bash\n-export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH\n-```\nRunning the demo application\n----------------------------\nType the following command to run the bare metal text recognition application ([src/demo_bare_metal.c](./src/demo_bare_metal.c)):\n+\n```bash\n./run_demo.sh\n```\n+\n+If you are not able to use Arm Virtual Hardware Amazon Machine Image(AMI) instance hosted by AWS/AWS China, specify argument --enable_FVP to 1 to make the application run on local Fixed Virtual Platforms (FVPs) executables.\n+\n+```bash\n+./run_demo.sh --enable_FVP 1\n+```\n+\nIf the Ethos(TM)-U platform and/or CMSIS have not been installed in /opt/arm/ethosu then\nthe locations for these can be specified as arguments to run_demo.sh, for example:\n@@ -65,13 +81,14 @@ the locations for these can be specified as arguments to run_demo.sh, for exampl\n--ethosu_platform_path /home/tvm-user/ethosu/core_platform\n```\n-This will:\n-- Download a PP-PicoDet text recognition model\n+With [run_demo.sh](./run_demo.sh) to run the demo application, it will:\n+- Set up running environment by installing the required prerequisites automatically if running in Arm Virtual Hardware Amazon AMI instance(not specify --enable_FVP to 1)\n+- Download a PP-PicoDet model\n- Use tvmc to compile the text recognition model for Cortex(R)-M55 CPU and CMSIS-NN\n- Create a C header file inputs.c containing the image data as a C array\n- Create a C header file outputs.c containing a C array where the output of inference will be stored\n- Build the demo application\n-- Run the demo application on a Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software\n+- Run the demo application on a Arm Virtual Hardware based on Arm(R) Corstone(TM)-300 software\n- The application will report the text on the image and the corresponding score.\nUsing your own image\n@@ -82,9 +99,9 @@ image to be converted into an array of bytes for consumption by the model.\nThe demo can be modified to use an image of your choice by changing the following line in run_demo.sh\n```bash\n-python3 ./convert_image.py ../../demo/000000014439_640x640.jpg\n+python3 ./convert_image.py path/to/image\n```\nModel description\n-----------------\n-In this demo, the model we used is based on [PP-PicoDet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/configs/picodet). Because of the excellent performance, PP-PicoDet are very suitable for deployment on mobile or CPU. And it is released by [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection).\n+In this demo, the model we used is based on [PP-PicoDet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/picodet). Because of the excellent performance, PP-PicoDet are very suitable for deployment on mobile or CPU. And it is released by [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection).\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_avh/README.md.bak",
"diff": "+<!--- Licensed to the Apache Software Foundation (ASF) under one -->\n+<!--- or more contributor license agreements. See the NOTICE file -->\n+<!--- distributed with this work for additional information -->\n+<!--- regarding copyright ownership. The ASF licenses this file -->\n+<!--- to you under the Apache License, Version 2.0 (the -->\n+<!--- \"License\"); you may not use this file except in compliance -->\n+<!--- with the License. You may obtain a copy of the License at -->\n+<!--- http://www.apache.org/licenses/LICENSE-2.0 -->\n+<!--- Unless required by applicable law or agreed to in writing, -->\n+<!--- software distributed under the License is distributed on an -->\n+<!--- \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->\n+<!--- KIND, either express or implied. See the License for the -->\n+<!--- specific language governing permissions and limitations -->\n+<!--- under the License. -->\n+Running PP-PicoDet via TVM on bare metal Arm(R) Cortex(R)-M55 CPU and CMSIS-NN\n+===============================================================\n+\n+This folder contains an example of how to use TVM to run a PP-PicoDet model\n+on bare metal Cortex(R)-M55 CPU and CMSIS-NN.\n+\n+Prerequisites\n+-------------\n+If the demo is run in the ci_cpu Docker container provided with TVM, then the following\n+software will already be installed.\n+\n+If the demo is not run in the ci_cpu Docker container, then you will need the following:\n+- Software required to build and run the demo (These can all be installed by running\n+ tvm/docker/install/ubuntu_install_ethosu_driver_stack.sh.)\n+ - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)\n+ - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)\n+ - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)\n+ - [Arm(R) Ethos(TM)-U NPU driver stack](https://review.mlplatform.org)\n+ - [CMSIS](https://github.com/ARM-software/CMSIS_5)\n+- The python libraries listed in the requirements.txt of this directory\n+ - These can be installed by running the following from the current directory:\n+ ```bash\n+ pip install -r ./requirements.txt\n+ ```\n+\n+You will also need TVM which can either be:\n+ - Built from source (see [Install from Source](https://tvm.apache.org/docs/install/from_source.html))\n+ - When building from source, the following need to be set in config.cmake:\n+ - set(USE_CMSISNN ON)\n+ - set(USE_MICRO ON)\n+ - set(USE_LLVM ON)\n+ - Installed from TLCPack(see [TLCPack](https://tlcpack.ai/))\n+\n+You will need to update your PATH environment variable to include the path to cmake 3.19.5 and the FVP.\n+For example if you've installed these in ```/opt/arm``` , then you would do the following:\n+```bash\n+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH\n+```\n+\n+Running the demo application\n+----------------------------\n+Type the following command to run the bare metal text recognition application ([src/demo_bare_metal.c](./src/demo_bare_metal.c)):\n+```bash\n+./run_demo.sh\n+```\n+If the Ethos(TM)-U platform and/or CMSIS have not been installed in /opt/arm/ethosu then\n+the locations for these can be specified as arguments to run_demo.sh, for example:\n+\n+```bash\n+./run_demo.sh --cmsis_path /home/tvm-user/cmsis \\\n+--ethosu_platform_path /home/tvm-user/ethosu/core_platform\n+```\n+\n+This will:\n+- Download a PP-PicoDet text recognition model\n+- Use tvmc to compile the text recognition model for Cortex(R)-M55 CPU and CMSIS-NN\n+- Create a C header file inputs.c containing the image data as a C array\n+- Create a C header file outputs.c containing a C array where the output of inference will be stored\n+- Build the demo application\n+- Run the demo application on a Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software\n+- The application will report the text on the image and the corresponding score.\n+\n+Using your own image\n+--------------------\n+The create_image.py script takes a single argument on the command line which is the path of the\n+image to be converted into an array of bytes for consumption by the model.\n+\n+The demo can be modified to use an image of your choice by changing the following line in run_demo.sh\n+\n+```bash\n+python3 ./convert_image.py ../../demo/000000014439_640x640.jpg\n+```\n+\n+Model description\n+-----------------\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_avh/configure_avh.sh",
"diff": "+#!/bin/bash\n+# Copyright (c) 2022 Arm Limited and Contributors. All rights reserved.\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+\n+set -e\n+set -u\n+set -o pipefail\n+\n+# Show usage\n+function show_usage() {\n+ cat <<EOF\n+Usage: Set up running environment by installing the required prerequisites.\n+-h, --help\n+ Display this help message.\n+EOF\n+}\n+\n+if [ \"$#\" -eq 1 ] && [ \"$1\" == \"--help\" -o \"$1\" == \"-h\" ]; then\n+ show_usage\n+ exit 0\n+elif [ \"$#\" -ge 1 ]; then\n+ show_usage\n+ exit 1\n+fi\n+\n+echo -e \"\\e[36mStart setting up running environment\\e[0m\"\n+\n+# Install CMSIS\n+echo -e \"\\e[36mStart installing CMSIS\\e[0m\"\n+CMSIS_PATH=\"/opt/arm/ethosu/cmsis\"\n+mkdir -p \"${CMSIS_PATH}\"\n+\n+CMSIS_SHA=\"977abe9849781a2e788b02282986480ff4e25ea6\"\n+CMSIS_SHASUM=\"86c88d9341439fbb78664f11f3f25bc9fda3cd7de89359324019a4d87d169939eea85b7fdbfa6ad03aa428c6b515ef2f8cd52299ce1959a5444d4ac305f934cc\"\n+CMSIS_URL=\"http://github.com/ARM-software/CMSIS_5/archive/${CMSIS_SHA}.tar.gz\"\n+DOWNLOAD_PATH=\"/tmp/${CMSIS_SHA}.tar.gz\"\n+\n+wget ${CMSIS_URL} -O \"${DOWNLOAD_PATH}\"\n+echo \"$CMSIS_SHASUM\" ${DOWNLOAD_PATH} | sha512sum -c\n+tar -xf \"${DOWNLOAD_PATH}\" -C \"${CMSIS_PATH}\" --strip-components=1\n+touch \"${CMSIS_PATH}\"/\"${CMSIS_SHA}\".sha\n+echo -e \"\\e[36mCMSIS Installation SUCCESS\\e[0m\"\n+\n+# Install Arm(R) Ethos(TM)-U NPU driver stack\n+echo -e \"\\e[36mStart installing Arm(R) Ethos(TM)-U NPU driver stack\\e[0m\"\n+git clone \"https://review.mlplatform.org/ml/ethos-u/ethos-u-core-platform\" /opt/arm/ethosu/core_platform\n+cd /opt/arm/ethosu/core_platform\n+git checkout tags/\"21.11\"\n+echo -e \"\\e[36mArm(R) Ethos(TM)-U Core Platform Installation SUCCESS\\e[0m\"\n+\n+# Install Arm(R) GNU Toolchain\n+echo -e \"\\e[36mStart installing Arm(R) GNU Toolchain\\e[0m\"\n+mkdir -p /opt/arm/gcc-arm-none-eabi\n+export gcc_arm_url='https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2?revision=ca0cbf9c-9de2-491c-ac48-898b5bbc0443&la=en&hash=68760A8AE66026BCF99F05AC017A6A50C6FD832A'\n+curl --retry 64 -sSL ${gcc_arm_url} | tar -C /opt/arm/gcc-arm-none-eabi --strip-components=1 -jx\n+export PATH=/opt/arm/gcc-arm-none-eabi/bin:$PATH\n+arm-none-eabi-gcc --version\n+arm-none-eabi-g++ --version\n+echo -e \"\\e[36mArm(R) Arm(R) GNU Toolchain Installation SUCCESS\\e[0m\"\n+\n+# Install TVM from TLCPack\n+echo -e \"\\e[36mStart installing TVM\\e[0m\"\n+pip install tlcpack-nightly -f https://tlcpack.ai/wheels\n+echo -e \"\\e[36mTVM Installation SUCCESS\\e[0m\"\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_avh/convert_image.py",
"new_path": "deploy/third_engine/demo_avh/convert_image.py",
"diff": "@@ -24,10 +24,10 @@ import math\nfrom PIL import Image\nimport numpy as np\n+\ndef resize_norm_img(img, image_shape, padding=True):\nimgC, imgH, imgW = image_shape\n- img = cv2.resize(\n- img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)\n+ img = cv2.resize(img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)\nimg = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\nimg = np.transpose(img, [2, 0, 1]) / 255\nimg = np.expand_dims(img, 0)\n@@ -47,9 +47,8 @@ def create_header_file(name, tensor_name, tensor_data, output_path):\nraw_path = file_path.with_suffix(\".h\").resolve()\nwith open(raw_path, \"a\") as header_file:\nheader_file.write(\n- \"\\n\"\n- + f\"const size_t {tensor_name}_len = {tensor_data.size};\\n\"\n- + f'__attribute__((section(\".data.tvm\"), aligned(16))) float {tensor_name}[] = '\n+ \"\\n\" + f\"const size_t {tensor_name}_len = {tensor_data.size};\\n\" +\n+ f'__attribute__((section(\".data.tvm\"), aligned(16))) float {tensor_name}[] = '\n)\nheader_file.write(\"{\")\n@@ -72,7 +71,9 @@ def create_headers(image_name):\n# # Add the batch dimension, as we are expecting 4-dimensional input: NCHW.\nimg_data = np.expand_dims(img_data, axis=0)\n+ if os.path.exists(\"./include/inputs.h\"):\nos.remove(\"./include/inputs.h\")\n+ if os.path.exists(\"./include/outputs.h\"):\nos.remove(\"./include/outputs.h\")\n# Create input header file\ncreate_header_file(\"inputs\", \"input\", img_data, \"./include\")\n@@ -82,15 +83,13 @@ def create_headers(image_name):\n\"outputs\",\n\"output0\",\noutput_data,\n- \"./include\",\n- )\n+ \"./include\", )\noutput_data = np.zeros([170000], np.float)\ncreate_header_file(\n\"outputs\",\n\"output1\",\noutput_data,\n- \"./include\",\n- )\n+ \"./include\", )\nif __name__ == \"__main__\":\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_avh/corstone300.ld",
"new_path": "deploy/third_engine/demo_avh/corstone300.ld",
"diff": "@@ -247,10 +247,10 @@ SECTIONS\n} > DTCM\n- .bss.NoInit :\n+ .bss.noinit (NOLOAD):\n{\n. = ALIGN(16);\n- *(.bss.NoInit)\n+ *(.bss.noinit.*)\n. = ALIGN(16);\n} > DDR AT > DDR\n"
},
{
"change_type": "ADD",
"old_path": "deploy/third_engine/demo_avh/image/000000014439_640x640.jpg",
"new_path": "deploy/third_engine/demo_avh/image/000000014439_640x640.jpg",
"diff": "Binary files /dev/null and b/deploy/third_engine/demo_avh/image/000000014439_640x640.jpg differ\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_avh/run_demo.sh",
"new_path": "deploy/third_engine/demo_avh/run_demo.sh",
"diff": "# KIND, either express or implied. See the License for the\n# specific language governing permissions and limitations\n# under the License.\n-export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH\nset -e\nset -u\nset -o pipefail\n@@ -34,9 +33,19 @@ Usage: run_demo.sh\nSet path to FVP.\n--cmake_path\nSet path to cmake.\n+--enable_FVP\n+ Set 1 to run application on local Fixed Virtual Platforms (FVPs) executables.\nEOF\n}\n+# Configure environment variables\n+FVP_enable=0\n+export PATH=/opt/arm/gcc-arm-none-eabi/bin:$PATH\n+\n+# Install python libraries\n+echo -e \"\\e[36mInstall python libraries\\e[0m\"\n+sudo pip install -r ./requirements.txt\n+\n# Parse arguments\nwhile (( $# )); do\ncase \"$1\" in\n@@ -93,6 +102,18 @@ while (( $# )); do\nfi\n;;\n+ --enable_FVP)\n+ if [ $# -gt 1 ] && [ \"$2\" == \"1\" -o \"$2\" == \"0\" ];\n+ then\n+ FVP_enable=\"$2\"\n+ shift 2\n+ else\n+ echo 'ERROR: --enable_FVP requires a right argument 1 or 0' >&2\n+ show_usage >&2\n+ exit 1\n+ fi\n+ ;;\n+\n-*|--*)\necho \"Error: Unknown flag: $1\" >&2\nshow_usage >&2\n@@ -101,6 +122,16 @@ while (( $# )); do\nesac\ndone\n+# Choose running environment: cloud(default) or local environment\n+Platform=\"VHT_Corstone_SSE-300_Ethos-U55\"\n+if [ $FVP_enable == \"1\" ]; then\n+ Platform=\"FVP_Corstone_SSE-300_Ethos-U55\"\n+ echo -e \"\\e[36mRun application on local Fixed Virtual Platforms (FVPs)\\e[0m\"\n+else\n+ if [ ! -d \"/opt/arm/\" ]; then\n+ sudo ./configure_avh.sh\n+ fi\n+fi\n# Directories\nscript_dir=\"$( cd \"$( dirname \"${BASH_SOURCE[0]}\" )\" &> /dev/null && pwd )\"\n@@ -110,6 +141,11 @@ make cleanall\nmkdir -p build\ncd build\n+# Get PaddlePaddle inference model\n+echo -e \"\\e[36mDownload PaddlePaddle inference model\\e[0m\"\n+wget https://bj.bcebos.com/v1/paddledet/deploy/Inference/picodet_s_320_coco_lcnet_no_nms.tar\n+tar -xf picodet_s_320_coco_lcnet_no_nms.tar\n+\n# Compile model for Arm(R) Cortex(R)-M55 CPU and CMSIS-NN\n# An alternative to using \"python3 -m tvm.driver.tvmc\" is to call\n# \"tvmc\" directly once TVM has been pip installed.\n@@ -123,7 +159,7 @@ python3 -m tvm.driver.tvmc compile --target=cmsis-nn,c \\\n--pass-config tir.usmp.enable=1 \\\n--pass-config tir.usmp.algorithm=hill_climb \\\n--pass-config tir.disable_storage_rewrite=1 \\\n- --pass-config tir.disable_vectorize=1 ../models/picodet_s_320_coco_lcnet_no_nms/model \\\n+ --pass-config tir.disable_vectorize=1 picodet_s_320_coco_lcnet_no_nms/model.pdmodel \\\n--output-format=mlf \\\n--model-format=paddle \\\n--module-name=picodet \\\n@@ -131,21 +167,18 @@ python3 -m tvm.driver.tvmc compile --target=cmsis-nn,c \\\n--output=picodet.tar\ntar -xf picodet.tar\n-\n# Create C header files\ncd ..\n-python3 ./convert_image.py ../../demo/000000014439_640x640.jpg\n+python3 ./convert_image.py ./image/000000014439_640x640.jpg\n# Build demo executable\n-echo \"Build demo executable...\"\ncd ${script_dir}\necho ${script_dir}\nmake\n-echo \"End build demo executable...\"\n-# Run demo executable on the FVP\n-FVP_Corstone_SSE-300_Ethos-U55 -C cpu0.CFGDTCMSZ=15 \\\n+# Run demo executable on the AVH\n+$Platform -C cpu0.CFGDTCMSZ=15 \\\n-C cpu0.CFGITCMSZ=15 -C mps3_board.uart0.out_file=\\\"-\\\" -C mps3_board.uart0.shutdown_tag=\\\"EXITTHESIM\\\" \\\n-C mps3_board.visualisation.disable-visualisation=1 -C mps3_board.telnetterminal0.start_telnet=0 \\\n-C mps3_board.telnetterminal1.start_telnet=0 -C mps3_board.telnetterminal2.start_telnet=0 -C mps3_board.telnetterminal5.start_telnet=0 \\\n-./build/demo\n+./build/demo --stat\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix avh demo (#6836)
|
499,298 |
05.09.2022 17:42:32
| -28,800 |
b127979f34289dd6b47303a9d06fd253b09a092a
|
[cherry-pick] add fuse_normalize for ppyoloe smalldet
|
[
{
"change_type": "MODIFY",
"old_path": "configs/smalldet/ppyoloe_crn_l_80e_sliced_visdrone_640_025.yml",
"new_path": "configs/smalldet/ppyoloe_crn_l_80e_sliced_visdrone_640_025.yml",
"diff": "@@ -20,6 +20,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nepoch: 80\nLearningRate:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/smalldet/ppyoloe_crn_l_80e_sliced_visdrone_640_025_slice_infer.yml",
"new_path": "configs/smalldet/ppyoloe_crn_l_80e_sliced_visdrone_640_025_slice_infer.yml",
"diff": "@@ -20,6 +20,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1 # only support bs=1 when slice infer\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nepoch: 80\nLearningRate:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/smalldet/ppyoloe_p2_crn_l_80e_sliced_DOTA_500_025.yml",
"new_path": "configs/smalldet/ppyoloe_p2_crn_l_80e_sliced_DOTA_500_025.yml",
"diff": "@@ -28,6 +28,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nepoch: 80\nLearningRate:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/smalldet/ppyoloe_p2_crn_l_80e_sliced_xview_400_025.yml",
"new_path": "configs/smalldet/ppyoloe_p2_crn_l_80e_sliced_xview_400_025.yml",
"diff": "@@ -28,6 +28,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nepoch: 80\nLearningRate:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/visdrone/ppyoloe_crn_l_80e_visdrone.yml",
"new_path": "configs/visdrone/ppyoloe_crn_l_80e_visdrone.yml",
"diff": "@@ -20,6 +20,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nepoch: 80\nLearningRate:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/visdrone/ppyoloe_crn_l_alpha_largesize_80e_visdrone.yml",
"new_path": "configs/visdrone/ppyoloe_crn_l_alpha_largesize_80e_visdrone.yml",
"diff": "@@ -55,3 +55,4 @@ TestReader:\n- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}\n- Permute: {}\nbatch_size: 1\n+ fuse_normalize: True\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/visdrone/ppyoloe_crn_l_p2_alpha_80e_visdrone.yml",
"new_path": "configs/visdrone/ppyoloe_crn_l_p2_alpha_80e_visdrone.yml",
"diff": "@@ -13,6 +13,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nLearningRate:\nbase_lr: 0.005\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/visdrone/ppyoloe_crn_l_p2_alpha_largesize_80e_visdrone.yml",
"new_path": "configs/visdrone/ppyoloe_crn_l_p2_alpha_largesize_80e_visdrone.yml",
"diff": "@@ -62,3 +62,4 @@ TestReader:\n- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}\n- Permute: {}\nbatch_size: 1\n+ fuse_normalize: True\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/visdrone/ppyoloe_crn_s_80e_visdrone.yml",
"new_path": "configs/visdrone/ppyoloe_crn_s_80e_visdrone.yml",
"diff": "@@ -20,6 +20,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nepoch: 80\nLearningRate:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/visdrone/ppyoloe_crn_s_p2_alpha_80e_visdrone.yml",
"new_path": "configs/visdrone/ppyoloe_crn_s_p2_alpha_80e_visdrone.yml",
"diff": "@@ -13,6 +13,10 @@ TrainReader:\nEvalReader:\nbatch_size: 1\n+TestReader:\n+ batch_size: 1\n+ fuse_normalize: True\n+\nLearningRate:\nbase_lr: 0.005\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] add fuse_normalize for ppyoloe smalldet (#6861)
|
499,298 |
05.09.2022 20:24:28
| -28,800 |
50d9662c20dae25b1faebd1ccfaccddd26c563d5
|
add ppyoloeplus visdrone
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "configs/visdrone/ppyoloe_plus_crn_l_largesize_80e_visdrone.yml",
"diff": "+_BASE_: [\n+ 'ppyoloe_crn_l_80e_visdrone.yml',\n+]\n+log_iter: 100\n+snapshot_epoch: 10\n+weights: output/ppyoloe_plus_crn_l_largesize_80e_visdrone/model_final\n+pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_l_80e_coco.pdparams\n+\n+\n+CSPResNet:\n+ use_alpha: True\n+\n+\n+LearningRate:\n+ base_lr: 0.0025\n+\n+\n+worker_num: 2\n+eval_height: &eval_height 1920\n+eval_width: &eval_width 1920\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\n+TrainReader:\n+ sample_transforms:\n+ - Decode: {}\n+ - RandomDistort: {}\n+ - RandomExpand: {fill_value: [123.675, 116.28, 103.53]}\n+ - RandomCrop: {}\n+ - RandomFlip: {}\n+ batch_transforms:\n+ - BatchRandomResize: {target_size: [1024, 1088, 1152, 1216, 1280, 1344, 1408, 1472, 1536, 1600, 1664, 1728, 1792, 1856, 1920], random_size: True, random_interp: True, keep_ratio: False}\n+ - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}\n+ - Permute: {}\n+ - PadGT: {}\n+ batch_size: 2\n+ shuffle: true\n+ drop_last: true\n+ use_shared_memory: true\n+ collate_batch: true\n+\n+EvalReader:\n+ sample_transforms:\n+ - Decode: {}\n+ - Resize: {target_size: *eval_size, keep_ratio: False, interp: 2}\n+ - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}\n+ - Permute: {}\n+ batch_size: 1\n+\n+TestReader:\n+ inputs_def:\n+ image_shape: [3, *eval_height, *eval_width]\n+ sample_transforms:\n+ - Decode: {}\n+ - Resize: {target_size: *eval_size, keep_ratio: False, interp: 2}\n+ - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}\n+ - Permute: {}\n+ batch_size: 1\n+ fuse_normalize: True\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -235,7 +235,7 @@ class Detector(object):\nimport sahi\nfrom sahi.slicing import slice_image\nexcept Exception as e:\n- logger.error(\n+ print(\n'sahi not found, plaese install sahi. '\n'for example: `pip install sahi`, see https://github.com/obss/sahi.'\n)\n@@ -251,6 +251,7 @@ class Detector(object):\noverlap_width_ratio=overlap_ratio[1])\nsub_img_num = len(slice_image_result)\nmerged_bboxs = []\n+ print('sub_img_num', sub_img_num)\nbatch_image_list = [\nslice_image_result.images[_ind] for _ind in range(sub_img_num)\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/visualize.py",
"new_path": "deploy/python/visualize.py",
"diff": "@@ -237,7 +237,7 @@ def visualize_pose(imgfile,\nimport matplotlib\nplt.switch_backend('agg')\nexcept Exception as e:\n- logger.error('Matplotlib not found, please install matplotlib.'\n+ print('Matplotlib not found, please install matplotlib.'\n'for example: `pip install matplotlib`.')\nraise e\nskeletons, scores = results['keypoint']\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/data/source/dataset.py",
"new_path": "ppdet/data/source/dataset.py",
"diff": "# limitations under the License.\nimport os\n+import copy\nimport numpy as np\n-\ntry:\nfrom collections.abc import Sequence\nexcept Exception:\n@@ -22,9 +22,11 @@ except Exception:\nfrom paddle.io import Dataset\nfrom ppdet.core.workspace import register, serializable\nfrom ppdet.utils.download import get_dataset_path\n-import copy\nfrom ppdet.data import source\n+from ppdet.utils.logger import setup_logger\n+logger = setup_logger(__name__)\n+\n@serializable\nclass DetDataset(Dataset):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add ppyoloeplus visdrone (#6863)
|
499,333 |
07.09.2022 12:12:31
| -28,800 |
06a719830cd39619f771f0ab1007bd86cbad63b6
|
update en doc and version check
|
[
{
"change_type": "MODIFY",
"old_path": "README.md",
"new_path": "README.md",
"diff": "-README_cn.md\n\\ No newline at end of file\n+README_en.md\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "docs/tutorials/INSTALL.md",
"new_path": "docs/tutorials/INSTALL.md",
"diff": "@@ -22,7 +22,7 @@ Dependency of PaddleDetection and PaddlePaddle:\n| PaddleDetection version | PaddlePaddle version | tips |\n| :----------------: | :---------------: | :-------: |\n-| develop | >= 2.2.2 | Dygraph mode is set as default |\n+| develop | develop | Dygraph mode is set as default |\n| release/2.5 | >= 2.2.2 | Dygraph mode is set as default |\n| release/2.4 | >= 2.2.2 | Dygraph mode is set as default |\n| release/2.3 | >= 2.2.0rc | Dygraph mode is set as default |\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/utils/check.py",
"new_path": "ppdet/utils/check.py",
"diff": "@@ -87,7 +87,7 @@ def check_gpu(use_gpu):\npass\n-def check_version(version='2.0'):\n+def check_version(version='2.2'):\n\"\"\"\nLog error and exit when the installed version of paddlepaddle is\nnot satisfied.\n@@ -100,8 +100,19 @@ def check_version(version='2.0'):\npaddle_version.major, paddle_version.minor, paddle_version.patch,\npaddle_version.rc\n]\n+\n+ # Paddledet develop version is only used on Paddle develop\n+ if version_installed == ['0', '0', '0', '0'] and version != 'develop':\n+ raise Exception(\n+ \"PaddlePaddle version {} or higher is required, and develop version is only used for PaddleDetection develop version!\".\n+ format(version))\n+\nif version_installed == ['0', '0', '0', '0']:\nreturn\n+\n+ if version == 'develop':\n+ raise Exception(\"PaddlePaddle develop version is required!\")\n+\nversion_split = version.split('.')\nlength = min(len(version_installed), len(version_split))\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update en doc and version check (#6876)
|
499,304 |
08.09.2022 21:16:40
| -28,800 |
3b36a65514c7b55461db98b02f95bd3216e4a637
|
fix distill error
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/slim/distill.py",
"new_path": "ppdet/slim/distill.py",
"diff": "@@ -262,7 +262,7 @@ class FGDFeatureLoss(nn.Layer):\nzeros_init = parameter_init(\"constant\", 0.0)\nif student_channels != teacher_channels:\n- self.align = nn.Conv2d(\n+ self.align = nn.Conv2D(\nstudent_channels,\nteacher_channels,\nkernel_size=1,\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix distill error (#6907)
|
499,339 |
08.09.2022 21:36:31
| -28,800 |
805420551109750bb68aeadbefb66928cb6a4f4a
|
[PPYOLOE] fix proj_conv in ptq bug
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"diff": "@@ -2,6 +2,7 @@ architecture: YOLOv3\nnorm_type: sync_bn\nuse_ema: true\nema_decay: 0.9998\n+ema_black_list: ['proj_conv.weight']\ncustom_black_list: ['reduce_mean']\nYOLOv3:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_plus_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_plus_crn.yml",
"diff": "@@ -2,6 +2,7 @@ architecture: YOLOv3\nnorm_type: sync_bn\nuse_ema: true\nema_decay: 0.9998\n+ema_black_list: ['proj_conv.weight']\ncustom_black_list: ['reduce_mean']\nYOLOv3:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/ppyoloe_crn_l_36e_coco_xpu.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_l_36e_coco_xpu.yml",
"diff": "@@ -26,6 +26,8 @@ architecture: YOLOv3\nnorm_type: sync_bn\nuse_ema: true\nema_decay: 0.9998\n+ema_black_list: ['proj_conv.weight']\n+custom_black_list: ['reduce_mean']\nYOLOv3:\nbackbone: CSPResNet\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/auto_compression/configs/ppyoloe_plus_m_qat_dis.yaml",
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_m_qat_dis.yaml",
"diff": "@@ -14,6 +14,7 @@ Distillation:\nQuantization:\nuse_pact: true\n+ onnx_format: True\nactivation_quantize_type: 'moving_average_abs_max'\nquantize_op_types:\n- conv2d\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/auto_compression/configs/ppyoloe_plus_reader.yml",
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_reader.yml",
"diff": "-\n-\nmetric: COCO\nnum_classes: 80\n@@ -23,6 +21,6 @@ EvalReader:\nsample_transforms:\n- Decode: {}\n- Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}\n- - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], is_scale: True}\n+ - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}\n- Permute: {}\nbatch_size: 4\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/auto_compression/configs/ppyoloe_plus_x_qat_dis.yaml",
"new_path": "deploy/auto_compression/configs/ppyoloe_plus_x_qat_dis.yaml",
"diff": "@@ -14,6 +14,7 @@ Distillation:\nQuantization:\nuse_pact: true\n+ onnx_format: True\nactivation_quantize_type: 'moving_average_abs_max'\nquantize_op_types:\n- conv2d\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/utils.py",
"new_path": "deploy/python/utils.py",
"diff": "@@ -108,7 +108,7 @@ def argsparser():\n\"calibration, trt_calib_mode need to set True.\")\nparser.add_argument(\n'--save_images',\n- type=bool,\n+ type=ast.literal_eval,\ndefault=True,\nhelp='Save visualization image results.')\nparser.add_argument(\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -169,13 +169,15 @@ class Trainer(object):\nself.use_ema = ('use_ema' in cfg and cfg['use_ema'])\nif self.use_ema:\nema_decay = self.cfg.get('ema_decay', 0.9998)\n- cycle_epoch = self.cfg.get('cycle_epoch', -1)\nema_decay_type = self.cfg.get('ema_decay_type', 'threshold')\n+ cycle_epoch = self.cfg.get('cycle_epoch', -1)\n+ ema_black_list = self.cfg.get('ema_black_list', None)\nself.ema = ModelEMA(\nself.model,\ndecay=ema_decay,\nema_decay_type=ema_decay_type,\n- cycle_epoch=cycle_epoch)\n+ cycle_epoch=cycle_epoch,\n+ ema_black_list=ema_black_list)\nself._nranks = dist.get_world_size()\nself._local_rank = dist.get_rank()\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/atss_assigner.py",
"new_path": "ppdet/modeling/assigners/atss_assigner.py",
"diff": "@@ -120,7 +120,7 @@ class ATSSAssigner(nn.Layer):\n# negative batch\nif num_max_boxes == 0:\nassigned_labels = paddle.full(\n- [batch_size, num_anchors], bg_index, dtype=gt_labels.dtype)\n+ [batch_size, num_anchors], bg_index, dtype='int32')\nassigned_bboxes = paddle.zeros([batch_size, num_anchors, 4])\nassigned_scores = paddle.zeros(\n[batch_size, num_anchors, self.num_classes])\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"new_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"diff": "@@ -86,7 +86,7 @@ class TaskAlignedAssigner(nn.Layer):\n# negative batch\nif num_max_boxes == 0:\nassigned_labels = paddle.full(\n- [batch_size, num_anchors], bg_index, dtype=gt_labels.dtype)\n+ [batch_size, num_anchors], bg_index, dtype='int32')\nassigned_bboxes = paddle.zeros([batch_size, num_anchors, 4])\nassigned_scores = paddle.zeros(\n[batch_size, num_anchors, num_classes])\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/ppyoloe_head.py",
"new_path": "ppdet/modeling/heads/ppyoloe_head.py",
"diff": "@@ -130,11 +130,10 @@ class PPYOLOEHead(nn.Layer):\nconstant_(reg_.weight)\nconstant_(reg_.bias, 1.0)\n- self.proj = paddle.linspace(0, self.reg_max, self.reg_max + 1)\n- self.proj_conv.weight.set_value(\n- self.proj.reshape([1, self.reg_max + 1, 1, 1]))\n+ proj = paddle.linspace(0, self.reg_max, self.reg_max + 1).reshape(\n+ [1, self.reg_max + 1, 1, 1])\n+ self.proj_conv.weight.set_value(proj)\nself.proj_conv.weight.stop_gradient = True\n-\nif self.eval_size:\nanchor_points, stride_tensor = self._generate_anchors()\nself.anchor_points = anchor_points\n@@ -200,15 +199,15 @@ class PPYOLOEHead(nn.Layer):\nfeat)\nreg_dist = self.pred_reg[i](self.stem_reg[i](feat, avg_feat))\nreg_dist = reg_dist.reshape([-1, 4, self.reg_max + 1, l]).transpose(\n- [0, 2, 1, 3])\n- reg_dist = self.proj_conv(F.softmax(reg_dist, axis=1))\n+ [0, 2, 3, 1])\n+ reg_dist = self.proj_conv(F.softmax(reg_dist, axis=1)).squeeze(1)\n# cls and reg\ncls_score = F.sigmoid(cls_logit)\ncls_score_list.append(cls_score.reshape([b, self.num_classes, l]))\n- reg_dist_list.append(reg_dist.reshape([b, 4, l]))\n+ reg_dist_list.append(reg_dist)\ncls_score_list = paddle.concat(cls_score_list, axis=-1)\n- reg_dist_list = paddle.concat(reg_dist_list, axis=-1)\n+ reg_dist_list = paddle.concat(reg_dist_list, axis=1)\nreturn cls_score_list, reg_dist_list, anchor_points, stride_tensor\n@@ -240,8 +239,8 @@ class PPYOLOEHead(nn.Layer):\ndef _bbox_decode(self, anchor_points, pred_dist):\nb, l, _ = get_static_shape(pred_dist)\n- pred_dist = F.softmax(pred_dist.reshape([b, l, 4, self.reg_max + 1\n- ])).matmul(self.proj)\n+ pred_dist = F.softmax(pred_dist.reshape([b, l, 4, self.reg_max + 1]))\n+ pred_dist = self.proj_conv(pred_dist.transpose([0, 3, 1, 2])).squeeze(1)\nreturn batch_distance2bbox(anchor_points, pred_dist)\ndef _bbox2distance(self, points, bbox):\n@@ -347,9 +346,8 @@ class PPYOLOEHead(nn.Layer):\nassigned_scores_sum = assigned_scores.sum()\nif paddle.distributed.get_world_size() > 1:\npaddle.distributed.all_reduce(assigned_scores_sum)\n- assigned_scores_sum = paddle.clip(\n- assigned_scores_sum / paddle.distributed.get_world_size(),\n- min=1)\n+ assigned_scores_sum /= paddle.distributed.get_world_size()\n+ assigned_scores_sum = paddle.clip(assigned_scores_sum, min=1.)\nloss_cls /= assigned_scores_sum\nloss_l1, loss_iou, loss_dfl = \\\n@@ -370,8 +368,7 @@ class PPYOLOEHead(nn.Layer):\ndef post_process(self, head_outs, scale_factor):\npred_scores, pred_dist, anchor_points, stride_tensor = head_outs\n- pred_bboxes = batch_distance2bbox(anchor_points,\n- pred_dist.transpose([0, 2, 1]))\n+ pred_bboxes = batch_distance2bbox(anchor_points, pred_dist)\npred_bboxes *= stride_tensor\nif self.exclude_post_process:\nreturn paddle.concat(\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer/ema.py",
"new_path": "ppdet/optimizer/ema.py",
"diff": "@@ -36,21 +36,30 @@ class ModelEMA(object):\nstep. Defaults is -1, which means not reset. Its function is to\nadd a regular effect to ema, which is set according to experience\nand is effective when the total training epoch is large.\n+ ema_black_list (set|list|tuple, optional): The custom EMA black_list.\n+ Blacklist of weight names that will not participate in EMA\n+ calculation. Default: None.\n\"\"\"\ndef __init__(self,\nmodel,\ndecay=0.9998,\nema_decay_type='threshold',\n- cycle_epoch=-1):\n+ cycle_epoch=-1,\n+ ema_black_list=None):\nself.step = 0\nself.epoch = 0\nself.decay = decay\n+ self.ema_decay_type = ema_decay_type\n+ self.cycle_epoch = cycle_epoch\n+ self.ema_black_list = self._match_ema_black_list(\n+ model.state_dict().keys(), ema_black_list)\nself.state_dict = dict()\nfor k, v in model.state_dict().items():\n+ if k in self.ema_black_list:\n+ self.state_dict[k] = v\n+ else:\nself.state_dict[k] = paddle.zeros_like(v)\n- self.ema_decay_type = ema_decay_type\n- self.cycle_epoch = cycle_epoch\nself._model_state = {\nk: weakref.ref(p)\n@@ -61,6 +70,9 @@ class ModelEMA(object):\nself.step = 0\nself.epoch = 0\nfor k, v in self.state_dict.items():\n+ if k in self.ema_black_list:\n+ self.state_dict[k] = v\n+ else:\nself.state_dict[k] = paddle.zeros_like(v)\ndef resume(self, state_dict, step=0):\n@@ -89,6 +101,7 @@ class ModelEMA(object):\n[v is not None for _, v in model_dict.items()]), 'python gc.'\nfor k, v in self.state_dict.items():\n+ if k not in self.ema_black_list:\nv = decay * v + (1 - decay) * model_dict[k]\nv.stop_gradient = True\nself.state_dict[k] = v\n@@ -99,6 +112,10 @@ class ModelEMA(object):\nreturn self.state_dict\nstate_dict = dict()\nfor k, v in self.state_dict.items():\n+ if k in self.ema_black_list:\n+ v.stop_gradient = True\n+ state_dict[k] = v\n+ else:\nif self.ema_decay_type != 'exponential':\nv = v / (1 - self._decay**self.step)\nv.stop_gradient = True\n@@ -108,3 +125,12 @@ class ModelEMA(object):\nself.reset()\nreturn state_dict\n+\n+ def _match_ema_black_list(self, weight_name, ema_black_list=None):\n+ out_list = set()\n+ if ema_black_list:\n+ for name in weight_name:\n+ for key in ema_black_list:\n+ if key in name:\n+ out_list.add(name)\n+ return out_list\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[PPYOLOE] fix proj_conv in ptq bug (#6908)
|
499,301 |
14.09.2022 11:30:22
| -28,800 |
7bd1ef4a473444048ab9b3a5568971f6fb76e5b6
|
repleace letterbox to resize_pad
|
[
{
"change_type": "MODIFY",
"old_path": "configs/faster_rcnn/_base_/faster_rcnn_swin_reader.yml",
"new_path": "configs/faster_rcnn/_base_/faster_rcnn_swin_reader.yml",
"diff": "@@ -33,7 +33,8 @@ TestReader:\nimage_shape: [-1, 3, 640, 640]\nsample_transforms:\n- Decode: {}\n- - LetterBoxResize: {target_size: 640}\n+ - Resize: {interp: 2, target_size: 640, keep_ratio: True}\n+ - Pad: {size: 640}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_size: 1\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/vitdet/_base_/reader.yml",
"new_path": "configs/vitdet/_base_/reader.yml",
"diff": "@@ -33,7 +33,8 @@ TestReader:\nimage_shape: [-1, 3, 640, 640]\nsample_transforms:\n- Decode: {}\n- - LetterBoxResize: {target_size: 640}\n+ - Resize: {interp: 2, target_size: 640, keep_ratio: True}\n+ - Pad: {size: 640}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_size: 1\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
repleace letterbox to resize_pad (#6934)
|
499,333 |
14.09.2022 20:08:00
| -28,800 |
6bc4f9437a459348b829cced7b5fe2e8001aa052
|
fix target when fg is empty
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/proposal_generator/target.py",
"new_path": "ppdet/modeling/proposal_generator/target.py",
"diff": "@@ -340,7 +340,7 @@ def generate_mask_target(gt_segms, rois, labels_int32, sampled_gt_inds,\n# generate fake roi if foreground is empty\nif fg_inds.numel() == 0:\nhas_fg = False\n- fg_inds = paddle.ones([1], dtype='int32')\n+ fg_inds = paddle.ones([1, 1], dtype='int64')\ninds_per_im = sampled_gt_inds[k]\ninds_per_im = paddle.gather(inds_per_im, fg_inds)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix target when fg is empty (#6937)
|
499,363 |
15.09.2022 13:45:26
| -28,800 |
92313771e21253ba345704ef5f71e06b3e41b857
|
add industry tutorial examples
|
[
{
"change_type": "MODIFY",
"old_path": "README_en.md",
"new_path": "README_en.md",
"diff": "@@ -464,6 +464,10 @@ Please refer to [docs](deploy/pipeline/README_en.md) for details.\n- [Visitor flow statistics based on FairMOT](https://aistudio.baidu.com/aistudio/projectdetail/2421822)\n+- [Guest analysis based on PP-Human](https://aistudio.baidu.com/aistudio/projectdetail/4537344)\n+\n+- [Fight recognition based on video classification](https://aistudio.baidu.com/aistudio/projectdetail/4512242)\n+\n- [More examples](./industrial_tutorial/README.md)\n## <img title=\"\" src=\"https://user-images.githubusercontent.com/48054808/157836473-1cf451fa-f01f-4148-ba68-b6d06d5da2f9.png\" alt=\"\" width=\"20\"> Applications\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add industry tutorial examples (#6945)
|
499,333 |
15.09.2022 15:41:57
| -28,800 |
92fb8b2e89db282cc425bc04ce20a942d1038062
|
fix deploy when bs=2
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -154,7 +154,7 @@ class Detector(object):\nnp_boxes_num = result['boxes_num']\nif sum(np_boxes_num) <= 0:\nprint('[WARNNING] No object detected.')\n- result = {'boxes': np.zeros([0, 6]), 'boxes_num': [0]}\n+ result = {'boxes': np.zeros([0, 6]), 'boxes_num': np_boxes_num}\nresult = {k: v for k, v in result.items() if v is not None}\nreturn result\n@@ -402,10 +402,8 @@ class Detector(object):\nself.pred_config.labels,\noutput_dir=self.output_dir,\nthreshold=self.threshold)\n-\nresults.append(result)\nprint('Test iter {}'.format(i))\n-\nresults = self.merge_batch_result(results)\nif save_results:\nPath(self.output_dir).mkdir(exist_ok=True)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix deploy when bs=2 (#6915)
|
499,298 |
23.09.2022 10:42:58
| -28,800 |
f8e57aebe7a7cec94a697445a9a425c0a4300957
|
[cherry-pick] update yoloseries doc
|
[
{
"change_type": "MODIFY",
"old_path": "docs/MODEL_ZOO_en.md",
"new_path": "docs/MODEL_ZOO_en.md",
"diff": "@@ -92,7 +92,19 @@ Please refer to[PP-YOLOE](https://github.com/PaddlePaddle/PaddleDetection/tree/r\n### YOLOX\n-Please refer to[YOLOX](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/yolox)\n+Please refer to[YOLOX](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/yolox)\n+\n+### YOLOv5\n+\n+Please refer to[YOLOv5](https://github.com/nemonameless/PaddleDetection_YOLOSeries/tree/develop/configs/yolov5)\n+\n+### YOLOv6\n+\n+Please refer to[YOLOv6](https://github.com/nemonameless/PaddleDetection_YOLOSeries/tree/develop/configs/yolov6)\n+\n+### YOLOv7\n+\n+Please refer to[YOLOv7](https://github.com/nemonameless/PaddleDetection_YOLOSeries/tree/develop/configs/yolov7)\n## Rotating frame detection\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] update yoloseries doc (#7006)
|
499,298 |
23.09.2022 10:58:36
| -28,800 |
54a5f0925f3e15c9d59f127f93be847c3cd48894
|
[cherry-pick] fix mot dataset link
|
[
{
"change_type": "MODIFY",
"old_path": "configs/mot/README_en.md",
"new_path": "configs/mot/README_en.md",
"diff": "@@ -86,18 +86,18 @@ PaddleDetection implement [JDE](https://github.com/Zhongdao/Towards-Realtime-MOT\n### Dataset Directory\nFirst, download the image_lists.zip using the following command, and unzip them into `PaddleDetection/dataset/mot`:\n```\n-wget https://dataset.bj.bcebos.com/mot/image_lists.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/image_lists.zip\n```\nThen, download the MIX dataset using the following command, and unzip them into `PaddleDetection/dataset/mot`:\n```\n-wget https://dataset.bj.bcebos.com/mot/MOT17.zip\n-wget https://dataset.bj.bcebos.com/mot/Caltech.zip\n-wget https://dataset.bj.bcebos.com/mot/CUHKSYSU.zip\n-wget https://dataset.bj.bcebos.com/mot/PRW.zip\n-wget https://dataset.bj.bcebos.com/mot/Cityscapes.zip\n-wget https://dataset.bj.bcebos.com/mot/ETHZ.zip\n-wget https://dataset.bj.bcebos.com/mot/MOT16.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/MOT17.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/Caltech.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/CUHKSYSU.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/PRW.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/Cityscapes.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/ETHZ.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/MOT16.zip\n```\nThe final directory is:\n"
},
{
"change_type": "MODIFY",
"old_path": "docs/tutorials/data/PrepareMOTDataSet_en.md",
"new_path": "docs/tutorials/data/PrepareMOTDataSet_en.md",
"diff": "@@ -20,18 +20,18 @@ PaddleDetection implement [JDE](https://github.com/Zhongdao/Towards-Realtime-MOT\n### Dataset Directory\nFirst, download the image_lists.zip using the following command, and unzip them into `PaddleDetection/dataset/mot`:\n```\n-wget https://dataset.bj.bcebos.com/mot/image_lists.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/image_lists.zip\n```\nThen, download the MIX dataset using the following command, and unzip them into `PaddleDetection/dataset/mot`:\n```\n-wget https://dataset.bj.bcebos.com/mot/MOT17.zip\n-wget https://dataset.bj.bcebos.com/mot/Caltech.zip\n-wget https://dataset.bj.bcebos.com/mot/CUHKSYSU.zip\n-wget https://dataset.bj.bcebos.com/mot/PRW.zip\n-wget https://dataset.bj.bcebos.com/mot/Cityscapes.zip\n-wget https://dataset.bj.bcebos.com/mot/ETHZ.zip\n-wget https://dataset.bj.bcebos.com/mot/MOT16.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/MOT17.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/Caltech.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/CUHKSYSU.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/PRW.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/Cityscapes.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/ETHZ.zip\n+wget https://bj.bcebos.com/v1/paddledet/data/mot/MOT16.zip\n```\nThe final directory is:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[cherry-pick] fix mot dataset link (#7003)
|
499,301 |
23.09.2022 11:19:39
| -28,800 |
7e19ac76273e30b64f5e34ce3a2f98e54abe4c8d
|
fix simota candidate_topk
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/simota_assigner.py",
"new_path": "ppdet/modeling/assigners/simota_assigner.py",
"diff": "@@ -115,7 +115,10 @@ class SimOTAAssigner(object):\ndef dynamic_k_matching(self, cost_matrix, pairwise_ious, num_gt):\nmatch_matrix = np.zeros_like(cost_matrix.numpy())\n# select candidate topk ious for dynamic-k calculation\n- topk_ious, _ = paddle.topk(pairwise_ious, self.candidate_topk, axis=0)\n+ topk_ious, _ = paddle.topk(\n+ pairwise_ious,\n+ min(self.candidate_topk, pairwise_ious.shape[0]),\n+ axis=0)\n# calculate dynamic k for each gt\ndynamic_ks = paddle.clip(topk_ious.sum(0).cast('int'), min=1)\nfor gt_idx in range(num_gt):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix simota candidate_topk (#6979) (#7001)
|
499,339 |
27.09.2022 16:34:00
| -28,800 |
94a6077b70700799f2db998a99cd06cff23375db
|
[dev] fix postprocess bug in deploy benchmark bug
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -152,7 +152,10 @@ class Detector(object):\ndef postprocess(self, inputs, result):\n# postprocess output of predictor\nnp_boxes_num = result['boxes_num']\n- if sum(np_boxes_num) <= 0:\n+ assert isinstance(np_boxes_num, np.ndarray), \\\n+ '`np_boxes_num` should be a `numpy.ndarray`'\n+\n+ if np_boxes_num.sum() <= 0:\nprint('[WARNNING] No object detected.')\nresult = {'boxes': np.zeros([0, 6]), 'boxes_num': np_boxes_num}\nresult = {k: v for k, v in result.items() if v is not None}\n@@ -188,7 +191,7 @@ class Detector(object):\nshape: [N, im_h, im_w]\n'''\n# model prediction\n- np_boxes, np_masks = None, None\n+ np_boxes_num, np_boxes, np_masks = np.array([0]), None, None\nfor i in range(repeats):\nself.predictor.run()\noutput_names = self.predictor.get_output_names()\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] fix postprocess bug in deploy benchmark bug (#7028) (#7031)
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.