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
|
|---|---|---|---|---|---|---|---|---|---|
499,301 |
31.12.2021 15:02:26
| -28,800 |
1a46f29fe1b42c6350cadf2b5ace66a217cdf145
|
params group
|
[
{
"change_type": "MODIFY",
"old_path": "configs/faster_rcnn/_base_/optimizer_swin_1x.yml",
"new_path": "configs/faster_rcnn/_base_/optimizer_swin_1x.yml",
"diff": "@@ -15,4 +15,8 @@ OptimizerBuilder:\noptimizer:\ntype: AdamW\nweight_decay: 0.05\n- without_weight_decay_params: ['absolute_pos_embed', 'relative_position_bias_table', 'norm']\n+\n+ param_groups:\n+ -\n+ params: ['absolute_pos_embed', 'relative_position_bias_table', 'norm']\n+ weight_decay: 0.\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer.py",
"diff": "@@ -249,21 +249,37 @@ class OptimizerBuilder():\noptim_args['weight_decay'] = regularization\nop = getattr(optimizer, optim_type)\n- if 'without_weight_decay_params' in optim_args:\n- keys = optim_args['without_weight_decay_params']\n- params = [{\n- 'params': [\n- p for n, p in model.named_parameters()\n- if any([k in n for k in keys])\n- ],\n- 'weight_decay': 0.\n- }, {\n- 'params': [\n- p for n, p in model.named_parameters()\n- if all([k not in n for k in keys])\n+ if 'param_groups' in optim_args:\n+ assert isinstance(optim_args['param_groups'], list), ''\n+\n+ param_groups = optim_args.pop('param_groups')\n+\n+ params, visited = [], []\n+ for group in param_groups:\n+ assert isinstance(group,\n+ dict) and 'params' in group and isinstance(\n+ group['params'], list), ''\n+ _params = {\n+ n: p\n+ for n, p in model.named_parameters()\n+ if any([k in n for k in group['params']])\n+ }\n+ _group = group.copy()\n+ _group.update({'params': list(_params.values())})\n+\n+ params.append(_group)\n+ visited.extend(list(_params.keys()))\n+\n+ ext_params = [\n+ p for n, p in model.named_parameters() if n not in visited\n]\n- }]\n- del optim_args['without_weight_decay_params']\n+\n+ if len(ext_params) < len(model.parameters()):\n+ params.append({'params': ext_params})\n+\n+ elif len(ext_params) > len(model.parameters()):\n+ raise RuntimeError\n+\nelse:\nparams = model.parameters()\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
params group (#4955)
|
499,301 |
04.01.2022 13:47:11
| -28,800 |
5f198f393dba3773b85d3b212e8d8df43812dccd
|
update test keep ratio same as eval
|
[
{
"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 +33,7 @@ TestReader:\nimage_shape: [1, 3, 640, 640]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [640, 640], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: [640, 640], keep_ratio: True}\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
|
update test keep ratio same as eval (#5051)
|
499,376 |
05.01.2022 14:06:14
| -28,800 |
2711a60b8773e1fdf579052dbb78988b3953a768
|
Change some config for det trainning performance
|
[
{
"change_type": "MODIFY",
"old_path": "configs/deformable_detr/_base_/deformable_detr_reader.yml",
"new_path": "configs/deformable_detr/_base_/deformable_detr_reader.yml",
"diff": "-worker_num: 0\n+worker_num: 2\nTrainReader:\nsample_transforms:\n- Decode: {}\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/mot/jde/_base_/jde_reader_1088x608.yml",
"new_path": "configs/mot/jde/_base_/jde_reader_1088x608.yml",
"diff": "-worker_num: 2\n+worker_num: 8\nTrainReader:\nsample_transforms:\n- Decode: {}\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/solov2/_base_/solov2_reader.yml",
"new_path": "configs/solov2/_base_/solov2_reader.yml",
"diff": "-worker_num: 2\n+worker_num: 8\nTrainReader:\nsample_transforms:\n- Decode: {}\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
Change some config for det trainning performance (#5055)
|
499,301 |
10.01.2022 21:41:25
| -28,800 |
e62c687486c0881759ffd49b736afb5ccaa3d717
|
ema update
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer.py",
"diff": "@@ -17,6 +17,7 @@ from __future__ import division\nfrom __future__ import print_function\nimport math\n+import weakref\nimport paddle\nimport paddle.nn as nn\n@@ -319,19 +320,31 @@ class ModelEMA(object):\nself.use_thres_step = use_thres_step\nself.cycle_epoch = cycle_epoch\n+ self._model_state = {\n+ k: weakref.ref(p)\n+ for k, p in model.state_dict().items()\n+ }\n+\ndef reset(self):\nself.step = 0\nself.epoch = 0\nfor k, v in self.state_dict.items():\nself.state_dict[k] = paddle.zeros_like(v)\n- def update(self, model):\n+ def update(self, model=None):\nif self.use_thres_step:\ndecay = min(self.decay, (1 + self.step) / (10 + self.step))\nelse:\ndecay = self.decay\nself._decay = decay\n+\n+ if model is not None:\nmodel_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+\nfor k, v in self.state_dict.items():\nv = decay * v + (1 - decay) * model_dict[k]\nv.stop_gradient = True\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
ema update (#5089)
|
499,304 |
13.01.2022 14:25:40
| -28,800 |
df55cb9b8cb3b3ef701738f82068e51528101748
|
update PicoDet and GFL post_process
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/_base_/picodet_esnet.yml",
"new_path": "configs/picodet/_base_/picodet_esnet.yml",
"diff": "architecture: PicoDet\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ESNet_x1_0_pretrained.pdparams\n-export_post_process: False # Whether post-processing is included in the network\n+export_post_process: False # Whether post-processing is included in the network when export model.\nPicoDet:\nbackbone: ESNet\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -631,9 +631,12 @@ class Trainer(object):\nim_shape = [image_shape[0], 2]\nscale_factor = [image_shape[0], 2]\n- export_post_process = self.cfg.get('export_post_process', False)\n- if hasattr(self.model, 'deploy') and not export_post_process:\n+ if hasattr(self.model, 'deploy'):\nself.model.deploy = True\n+ export_post_process = self.cfg.get('export_post_process', False)\n+ if hasattr(self.model, 'export_post_process'):\n+ self.model.export_post_process = export_post_process\n+ image_shape = [None] + image_shape[1:]\nif hasattr(self.model, 'fuse_norm'):\nself.model.fuse_norm = self.cfg['TestReader'].get('fuse_normalize',\nFalse)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/architectures/picodet.py",
"new_path": "ppdet/modeling/architectures/picodet.py",
"diff": "@@ -41,7 +41,7 @@ class PicoDet(BaseArch):\nself.backbone = backbone\nself.neck = neck\nself.head = head\n- self.deploy = False\n+ self.export_post_process = True\n@classmethod\ndef from_config(cls, cfg, *args, **kwargs):\n@@ -62,8 +62,8 @@ class PicoDet(BaseArch):\ndef _forward(self):\nbody_feats = self.backbone(self.inputs)\nfpn_feats = self.neck(body_feats)\n- head_outs = self.head(fpn_feats, self.deploy)\n- if self.training or self.deploy:\n+ head_outs = self.head(fpn_feats, self.export_post_process)\n+ if self.training or not self.export_post_process:\nreturn head_outs, None\nelse:\nim_shape = self.inputs['im_shape']\n@@ -83,7 +83,7 @@ class PicoDet(BaseArch):\nreturn loss\ndef get_pred(self):\n- if self.deploy:\n+ if not self.export_post_process:\nreturn {'picodet': self._forward()[0]}\nelse:\nbbox_pred, bbox_num = self._forward()\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/bbox_utils.py",
"new_path": "ppdet/modeling/bbox_utils.py",
"diff": "@@ -756,20 +756,22 @@ def bbox_center(boxes):\ndef batch_distance2bbox(points, distance, max_shapes=None):\n\"\"\"Decode distance prediction to bounding box for batch.\nArgs:\n- points (Tensor): [B, ..., 2]\n- distance (Tensor): [B, ..., 4]\n- max_shapes (tuple): [B, 2], \"h,w\" format, Shape of the image.\n+ points (Tensor): [B, ..., 2], \"xy\" format\n+ distance (Tensor): [B, ..., 4], \"ltrb\" format\n+ max_shapes (Tensor): [B, 2], \"h,w\" format, Shape of the image.\nReturns:\n- Tensor: Decoded bboxes.\n+ Tensor: Decoded bboxes, \"x1y1x2y2\" format.\n\"\"\"\n- x1 = points[..., 0] - distance[..., 0]\n- y1 = points[..., 1] - distance[..., 1]\n- x2 = points[..., 0] + distance[..., 2]\n- y2 = points[..., 1] + distance[..., 3]\n+ lt, rb = paddle.split(distance, 2, -1)\n+ x1y1 = points - lt\n+ x2y2 = points + rb\n+ out_bbox = paddle.concat([x1y1, x2y2], -1)\nif max_shapes is not None:\n- for i, max_shape in enumerate(max_shapes):\n- x1[i] = x1[i].clip(min=0, max=max_shape[1])\n- y1[i] = y1[i].clip(min=0, max=max_shape[0])\n- x2[i] = x2[i].clip(min=0, max=max_shape[1])\n- y2[i] = y2[i].clip(min=0, max=max_shape[0])\n- return paddle.stack([x1, y1, x2, y2], -1)\n+ max_shapes = max_shapes.flip(-1).tile([1, 2])\n+ delta_dim = out_bbox.ndim - max_shapes.ndim\n+ for _ in range(delta_dim):\n+ max_shapes.unsqueeze_(1)\n+ out_bbox = paddle.where(out_bbox < max_shapes, out_bbox, max_shapes)\n+ out_bbox = paddle.where(out_bbox > 0, out_bbox,\n+ paddle.zeros_like(out_bbox))\n+ return out_bbox\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/gfl_head.py",
"new_path": "ppdet/modeling/heads/gfl_head.py",
"diff": "@@ -29,7 +29,7 @@ from paddle.nn.initializer import Normal, Constant\nfrom ppdet.core.workspace import register\nfrom ppdet.modeling.layers import ConvNormLayer\n-from ppdet.modeling.bbox_utils import distance2bbox, bbox2distance\n+from ppdet.modeling.bbox_utils import distance2bbox, bbox2distance, batch_distance2bbox\nfrom ppdet.data.transform.atss_assigner import bbox_overlaps\n@@ -241,18 +241,34 @@ class GFLHead(nn.Layer):\n), \"The size of fpn_feats is not equal to size of fpn_stride\"\ncls_logits_list = []\nbboxes_reg_list = []\n- for scale_reg, fpn_feat in zip(self.scales_regs, fpn_feats):\n+ for stride, scale_reg, fpn_feat in zip(self.fpn_stride,\n+ self.scales_regs, fpn_feats):\nconv_cls_feat, conv_reg_feat = self.conv_feat(fpn_feat)\n- cls_logits = self.gfl_head_cls(conv_cls_feat)\n- bbox_reg = scale_reg(self.gfl_head_reg(conv_reg_feat))\n+ cls_score = self.gfl_head_cls(conv_cls_feat)\n+ bbox_pred = scale_reg(self.gfl_head_reg(conv_reg_feat))\nif self.dgqp_module:\n- quality_score = self.dgqp_module(bbox_reg)\n- cls_logits = F.sigmoid(cls_logits) * quality_score\n+ quality_score = self.dgqp_module(bbox_pred)\n+ cls_score = F.sigmoid(cls_score) * quality_score\nif not self.training:\n- cls_logits = F.sigmoid(cls_logits.transpose([0, 2, 3, 1]))\n- bbox_reg = bbox_reg.transpose([0, 2, 3, 1])\n- cls_logits_list.append(cls_logits)\n- bboxes_reg_list.append(bbox_reg)\n+ cls_score = F.sigmoid(cls_score.transpose([0, 2, 3, 1]))\n+ bbox_pred = bbox_pred.transpose([0, 2, 3, 1])\n+ b, cell_h, cell_w, _ = paddle.shape(cls_score)\n+ y, x = self.get_single_level_center_point(\n+ [cell_h, cell_w], stride, cell_offset=self.cell_offset)\n+ center_points = paddle.stack([x, y], axis=-1)\n+ cls_score = cls_score.reshape([b, -1, self.cls_out_channels])\n+ bbox_pred = self.distribution_project(bbox_pred) * stride\n+ bbox_pred = bbox_pred.reshape([b, cell_h * cell_w, 4])\n+\n+ # NOTE: If keep_ratio=False and image shape value that\n+ # multiples of 32, distance2bbox not set max_shapes parameter\n+ # to speed up model prediction. If need to set max_shapes,\n+ # please use inputs['im_shape'].\n+ bbox_pred = batch_distance2bbox(\n+ center_points, bbox_pred, max_shapes=None)\n+\n+ cls_logits_list.append(cls_score)\n+ bboxes_reg_list.append(bbox_pred)\nreturn (cls_logits_list, bboxes_reg_list)\n@@ -410,71 +426,15 @@ class GFLHead(nn.Layer):\nx = x.flatten()\nreturn y, x\n- def get_bboxes_single(self,\n- cls_scores,\n- bbox_preds,\n- img_shape,\n- scale_factor,\n- rescale=True,\n- cell_offset=0):\n- assert len(cls_scores) == len(bbox_preds)\n- mlvl_bboxes = []\n- mlvl_scores = []\n- for stride, cls_score, bbox_pred in zip(self.fpn_stride, cls_scores,\n- bbox_preds):\n- featmap_size = [\n- paddle.shape(cls_score)[0], paddle.shape(cls_score)[1]\n- ]\n- y, x = self.get_single_level_center_point(\n- featmap_size, stride, cell_offset=cell_offset)\n- center_points = paddle.stack([x, y], axis=-1)\n- scores = cls_score.reshape([-1, self.cls_out_channels])\n- bbox_pred = self.distribution_project(bbox_pred) * stride\n-\n- if scores.shape[0] > self.nms_pre:\n- max_scores = scores.max(axis=1)\n- _, topk_inds = max_scores.topk(self.nms_pre)\n- center_points = center_points.gather(topk_inds)\n- bbox_pred = bbox_pred.gather(topk_inds)\n- scores = scores.gather(topk_inds)\n-\n- bboxes = distance2bbox(\n- center_points, bbox_pred, max_shape=img_shape)\n- mlvl_bboxes.append(bboxes)\n- mlvl_scores.append(scores)\n- mlvl_bboxes = paddle.concat(mlvl_bboxes)\n- if rescale:\n- # [h_scale, w_scale] to [w_scale, h_scale, w_scale, h_scale]\n- im_scale = paddle.concat([scale_factor[::-1], scale_factor[::-1]])\n- mlvl_bboxes /= im_scale\n- mlvl_scores = paddle.concat(mlvl_scores)\n- mlvl_scores = mlvl_scores.transpose([1, 0])\n- return mlvl_bboxes, mlvl_scores\n-\n- def decode(self, cls_scores, bbox_preds, im_shape, scale_factor,\n- cell_offset):\n- batch_bboxes = []\n- batch_scores = []\n- for img_id in range(cls_scores[0].shape[0]):\n- num_levels = len(cls_scores)\n- cls_score_list = [cls_scores[i][img_id] for i in range(num_levels)]\n- bbox_pred_list = [bbox_preds[i][img_id] for i in range(num_levels)]\n- bboxes, scores = self.get_bboxes_single(\n- cls_score_list,\n- bbox_pred_list,\n- im_shape[img_id],\n- scale_factor[img_id],\n- cell_offset=cell_offset)\n- batch_bboxes.append(bboxes)\n- batch_scores.append(scores)\n- batch_bboxes = paddle.stack(batch_bboxes, axis=0)\n- batch_scores = paddle.stack(batch_scores, axis=0)\n-\n- return batch_bboxes, batch_scores\n-\ndef post_process(self, gfl_head_outs, im_shape, scale_factor):\ncls_scores, bboxes_reg = gfl_head_outs\n- bboxes, score = self.decode(cls_scores, bboxes_reg, im_shape,\n- scale_factor, self.cell_offset)\n- bbox_pred, bbox_num, _ = self.nms(bboxes, score)\n+ bboxes = paddle.concat(bboxes_reg, axis=1)\n+ # rescale: [h_scale, w_scale] -> [w_scale, h_scale, w_scale, h_scale]\n+ im_scale = paddle.concat(\n+ [scale_factor[:, ::-1], scale_factor[:, ::-1]],\n+ axis=-1).unsqueeze(1)\n+ bboxes /= im_scale\n+ mlvl_scores = paddle.concat(cls_scores, axis=1)\n+ mlvl_scores = mlvl_scores.transpose([0, 2, 1])\n+ bbox_pred, bbox_num, _ = self.nms(bboxes, mlvl_scores)\nreturn bbox_pred, bbox_num\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/pico_head.py",
"new_path": "ppdet/modeling/heads/pico_head.py",
"diff": "@@ -26,6 +26,7 @@ from paddle.nn.initializer import Normal, Constant\nfrom ppdet.core.workspace import register\nfrom ppdet.modeling.layers import ConvNormLayer\n+from ppdet.modeling.bbox_utils import batch_distance2bbox\nfrom .simota_head import OTAVFLHead\n@@ -238,7 +239,7 @@ class PicoHead(OTAVFLHead):\nbias_attr=ParamAttr(initializer=Constant(value=0))))\nself.head_reg_list.append(head_reg)\n- def forward(self, fpn_feats, deploy=False):\n+ def forward(self, fpn_feats, export_post_process=True):\nassert len(fpn_feats) == len(\nself.fpn_stride\n), \"The size of fpn_feats is not equal to size of fpn_stride\"\n@@ -260,7 +261,7 @@ class PicoHead(OTAVFLHead):\nquality_score = self.dgqp_module(bbox_pred)\ncls_score = F.sigmoid(cls_score) * quality_score\n- if deploy:\n+ if not export_post_process:\n# Now only supports batch size = 1 in deploy\n# TODO(ygh): support batch size > 1\ncls_score = F.sigmoid(cls_score).reshape(\n@@ -270,6 +271,21 @@ class PicoHead(OTAVFLHead):\nelif not self.training:\ncls_score = F.sigmoid(cls_score.transpose([0, 2, 3, 1]))\nbbox_pred = bbox_pred.transpose([0, 2, 3, 1])\n+ stride = self.fpn_stride[i]\n+ b, cell_h, cell_w, _ = paddle.shape(cls_score)\n+ y, x = self.get_single_level_center_point(\n+ [cell_h, cell_w], stride, cell_offset=self.cell_offset)\n+ center_points = paddle.stack([x, y], axis=-1)\n+ cls_score = cls_score.reshape([b, -1, self.cls_out_channels])\n+ bbox_pred = self.distribution_project(bbox_pred) * stride\n+ bbox_pred = bbox_pred.reshape([b, cell_h * cell_w, 4])\n+\n+ # NOTE: If keep_ratio=False and image shape value that\n+ # multiples of 32, distance2bbox not set max_shapes parameter\n+ # to speed up model prediction. If need to set max_shapes,\n+ # please use inputs['im_shape'].\n+ bbox_pred = batch_distance2bbox(\n+ center_points, bbox_pred, max_shapes=None)\ncls_logits_list.append(cls_score)\nbboxes_reg_list.append(bbox_pred)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update PicoDet and GFL post_process (#5101)
|
499,340 |
19.01.2022 11:05:46
| -28,800 |
3233ef340b0d387641d0a6e8c173bdae14552c12
|
support sparse evaluation
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -130,6 +130,7 @@ class Trainer(object):\nself.lr = create('LearningRate')(steps_per_epoch)\nself.optimizer = create('OptimizerBuilder')(self.lr, self.model)\n+ # Unstructured pruner is only enabled in the train mode.\nif self.cfg.get('unstructured_prune'):\nself.pruner = create('UnstructuredPruner')(self.model,\nsteps_per_epoch)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
support sparse evaluation (#5122)
|
499,298 |
28.01.2022 17:38:16
| -28,800 |
6ee18c2b3964f50ea5a850cec224f0c06f01a403
|
fix reid resnet bn export, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/reid/resnet.py",
"new_path": "ppdet/modeling/reid/resnet.py",
"diff": "@@ -55,7 +55,7 @@ class ConvBNLayer(nn.Layer):\nbias_attr=False,\ndata_format=data_format)\n- self._batch_norm = nn.BatchNorm2D(num_filters, data_layout=data_format)\n+ self._batch_norm = nn.BatchNorm2D(num_filters)\nself.act = act\ndef forward(self, inputs):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix reid resnet bn export, test=document_fix (#5162)
|
499,339 |
29.01.2022 19:08:14
| -28,800 |
0a3d768ce3464fca945ba58f0742fbe003930ec7
|
[dev] update assigner and tood_head
|
[
{
"change_type": "MODIFY",
"old_path": "configs/tood/_base_/tood_reader.yml",
"new_path": "configs/tood/_base_/tood_reader.yml",
"diff": "@@ -8,10 +8,11 @@ TrainReader:\n- Permute: {}\nbatch_transforms:\n- PadBatch: {pad_to_stride: 32}\n+ - PadGT: {}\nbatch_size: 4\nshuffle: true\ndrop_last: true\n- collate_batch: false\n+ collate_batch: true\nuse_shared_memory: true\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/data/transform/batch_operators.py",
"new_path": "ppdet/data/transform/batch_operators.py",
"diff": "@@ -47,6 +47,7 @@ __all__ = [\n'PadMaskBatch',\n'Gt2GFLTarget',\n'Gt2CenterNetTarget',\n+ 'PadGT',\n]\n@@ -72,13 +73,15 @@ class PadBatch(BaseOperator):\ncoarsest_stride = self.pad_to_stride\n# multi scale input is nested list\n- if isinstance(samples, typing.Sequence) and len(samples) > 0 and isinstance(samples[0], typing.Sequence):\n+ if isinstance(samples,\n+ typing.Sequence) and len(samples) > 0 and isinstance(\n+ samples[0], typing.Sequence):\ninner_samples = samples[0]\nelse:\ninner_samples = samples\n- max_shape = np.array([data['image'].shape for data in inner_samples]).max(\n- axis=0)\n+ max_shape = np.array(\n+ [data['image'].shape for data in inner_samples]).max(axis=0)\nif coarsest_stride > 0:\nmax_shape[1] = int(\nnp.ceil(max_shape[1] / coarsest_stride) * coarsest_stride)\n@@ -1066,3 +1069,56 @@ class Gt2CenterNetTarget(BaseOperator):\nsample['size'] = wh\nsample['offset'] = reg\nreturn sample\n+\n+\n+@register_op\n+class PadGT(BaseOperator):\n+ \"\"\"\n+ Pad 0 to `gt_class`, `gt_bbox`, `gt_score`...\n+ The num_max_boxes is the largest for batch.\n+ Args:\n+ return_gt_mask (bool): If true, return `pad_gt_mask`,\n+ 1 means bbox, 0 means no bbox.\n+ \"\"\"\n+\n+ def __init__(self, return_gt_mask=True):\n+ super(PadGT, self).__init__()\n+ self.return_gt_mask = return_gt_mask\n+\n+ def __call__(self, samples, context=None):\n+ num_max_boxes = max([len(s['gt_bbox']) for s in samples])\n+ for sample in samples:\n+ if self.return_gt_mask:\n+ sample['pad_gt_mask'] = np.zeros(\n+ (num_max_boxes, 1), dtype=np.float32)\n+ if num_max_boxes == 0:\n+ continue\n+\n+ num_gt = len(sample['gt_bbox'])\n+ pad_gt_class = np.zeros((num_max_boxes, 1), dtype=np.int32)\n+ pad_gt_bbox = np.zeros((num_max_boxes, 4), dtype=np.float32)\n+ if num_gt > 0:\n+ pad_gt_class[:num_gt] = sample['gt_class']\n+ pad_gt_bbox[:num_gt] = sample['gt_bbox']\n+ sample['gt_class'] = pad_gt_class\n+ sample['gt_bbox'] = pad_gt_bbox\n+ # pad_gt_mask\n+ if 'pad_gt_mask' in sample:\n+ sample['pad_gt_mask'][:num_gt] = 1\n+ # gt_score\n+ if 'gt_score' in sample:\n+ pad_gt_score = np.zeros((num_max_boxes, 1), dtype=np.float32)\n+ if num_gt > 0:\n+ pad_gt_score[:num_gt] = sample['gt_score']\n+ sample['gt_score'] = pad_gt_score\n+ if 'is_crowd' in sample:\n+ pad_is_crowd = np.zeros((num_max_boxes, 1), dtype=np.int32)\n+ if num_gt > 0:\n+ pad_is_crowd[:num_gt] = sample['is_crowd']\n+ sample['is_crowd'] = pad_is_crowd\n+ if 'difficult' in sample:\n+ pad_diff = np.zeros((num_max_boxes, 1), dtype=np.int32)\n+ if num_gt > 0:\n+ pad_diff[:num_gt] = sample['difficult']\n+ sample['difficult'] = pad_diff\n+ return samples\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/atss_assigner.py",
"new_path": "ppdet/modeling/assigners/atss_assigner.py",
"diff": "@@ -23,10 +23,13 @@ import paddle.nn.functional as F\nfrom ppdet.core.workspace import register\nfrom ..ops import iou_similarity\n+from ..bbox_utils import iou_similarity as batch_iou_similarity\nfrom ..bbox_utils import bbox_center\n-from .utils import (pad_gt, check_points_inside_bboxes, compute_max_iou_anchor,\n+from .utils import (check_points_inside_bboxes, compute_max_iou_anchor,\ncompute_max_iou_gt)\n+__all__ = ['ATSSAssigner']\n+\n@register\nclass ATSSAssigner(nn.Layer):\n@@ -77,8 +80,10 @@ class ATSSAssigner(nn.Layer):\nnum_anchors_list,\ngt_labels,\ngt_bboxes,\n+ pad_gt_mask,\nbg_index,\n- gt_scores=None):\n+ gt_scores=None,\n+ pred_bboxes=None):\nr\"\"\"This code is based on\nhttps://github.com/fcjian/TOOD/blob/master/mmdet/core/bbox/assigners/atss_assigner.py\n@@ -99,18 +104,18 @@ class ATSSAssigner(nn.Layer):\nanchor_bboxes (Tensor, float32): pre-defined anchors, shape(L, 4),\n\"xmin, xmax, ymin, ymax\" format\nnum_anchors_list (List): num of anchors in each level\n- gt_labels (Tensor|List[Tensor], int64): Label of gt_bboxes, shape(B, n, 1)\n- gt_bboxes (Tensor|List[Tensor], float32): Ground truth bboxes, shape(B, n, 4)\n+ gt_labels (Tensor, int64|int32): Label of gt_bboxes, shape(B, n, 1)\n+ gt_bboxes (Tensor, float32): Ground truth bboxes, shape(B, n, 4)\n+ pad_gt_mask (Tensor, float32): 1 means bbox, 0 means no bbox, shape(B, n, 1)\nbg_index (int): background index\n- gt_scores (Tensor|List[Tensor]|None, float32) Score of gt_bboxes,\n+ gt_scores (Tensor|None, float32) Score of gt_bboxes,\nshape(B, n, 1), if None, then it will initialize with one_hot label\n+ pred_bboxes (Tensor, float32, optional): predicted bounding boxes, shape(B, L, 4)\nReturns:\nassigned_labels (Tensor): (B, L)\nassigned_bboxes (Tensor): (B, L, 4)\n- assigned_scores (Tensor): (B, L, C)\n+ assigned_scores (Tensor): (B, L, C), if pred_bboxes is not None, then output ious\n\"\"\"\n- gt_labels, gt_bboxes, pad_gt_scores, pad_gt_mask = pad_gt(\n- gt_labels, gt_bboxes, gt_scores)\nassert gt_labels.ndim == gt_bboxes.ndim and \\\ngt_bboxes.ndim == 3\n@@ -198,9 +203,14 @@ class ATSSAssigner(nn.Layer):\nassigned_bboxes = assigned_bboxes.reshape([batch_size, num_anchors, 4])\nassigned_scores = F.one_hot(assigned_labels, self.num_classes)\n- if gt_scores is not None:\n+ if pred_bboxes is not None:\n+ # assigned iou\n+ ious = batch_iou_similarity(gt_bboxes, pred_bboxes) * mask_positive\n+ ious = ious.max(axis=-2).unsqueeze(-1)\n+ assigned_scores *= ious\n+ elif gt_scores is not None:\ngather_scores = paddle.gather(\n- pad_gt_scores.flatten(), assigned_gt_index.flatten(), axis=0)\n+ gt_scores.flatten(), assigned_gt_index.flatten(), axis=0)\ngather_scores = gather_scores.reshape([batch_size, num_anchors])\ngather_scores = paddle.where(mask_positive_sum > 0, gather_scores,\npaddle.zeros_like(gather_scores))\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"new_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"diff": "@@ -22,9 +22,11 @@ import paddle.nn.functional as F\nfrom ppdet.core.workspace import register\nfrom ..bbox_utils import iou_similarity\n-from .utils import (pad_gt, gather_topk_anchors, check_points_inside_bboxes,\n+from .utils import (gather_topk_anchors, check_points_inside_bboxes,\ncompute_max_iou_anchor)\n+__all__ = ['TaskAlignedAssigner']\n+\n@register\nclass TaskAlignedAssigner(nn.Layer):\n@@ -43,8 +45,10 @@ class TaskAlignedAssigner(nn.Layer):\npred_scores,\npred_bboxes,\nanchor_points,\n+ num_anchors_list,\ngt_labels,\ngt_bboxes,\n+ pad_gt_mask,\nbg_index,\ngt_scores=None):\nr\"\"\"This code is based on\n@@ -61,20 +65,18 @@ class TaskAlignedAssigner(nn.Layer):\npred_scores (Tensor, float32): predicted class probability, shape(B, L, C)\npred_bboxes (Tensor, float32): predicted bounding boxes, shape(B, L, 4)\nanchor_points (Tensor, float32): pre-defined anchors, shape(L, 2), \"cxcy\" format\n- gt_labels (Tensor|List[Tensor], int64): Label of gt_bboxes, shape(B, n, 1)\n- gt_bboxes (Tensor|List[Tensor], float32): Ground truth bboxes, shape(B, n, 4)\n+ num_anchors_list (List): num of anchors in each level, shape(L)\n+ gt_labels (Tensor, int64|int32): Label of gt_bboxes, shape(B, n, 1)\n+ gt_bboxes (Tensor, float32): Ground truth bboxes, shape(B, n, 4)\n+ pad_gt_mask (Tensor, float32): 1 means bbox, 0 means no bbox, shape(B, n, 1)\nbg_index (int): background index\n- gt_scores (Tensor|List[Tensor]|None, float32) Score of gt_bboxes,\n- shape(B, n, 1), if None, then it will initialize with one_hot label\n+ gt_scores (Tensor|None, float32) Score of gt_bboxes, shape(B, n, 1)\nReturns:\nassigned_labels (Tensor): (B, L)\nassigned_bboxes (Tensor): (B, L, 4)\nassigned_scores (Tensor): (B, L, C)\n\"\"\"\nassert pred_scores.ndim == pred_bboxes.ndim\n-\n- gt_labels, gt_bboxes, pad_gt_scores, pad_gt_mask = pad_gt(\n- gt_labels, gt_bboxes, gt_scores)\nassert gt_labels.ndim == gt_bboxes.ndim and \\\ngt_bboxes.ndim == 3\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/tood_head.py",
"new_path": "ppdet/modeling/heads/tood_head.py",
"diff": "@@ -286,9 +286,11 @@ class TOODHead(nn.Layer):\nreturn loss\ndef get_loss(self, head_outs, gt_meta):\n- pred_scores, pred_bboxes, anchors, num_anchors_list, stride_tensor_list = head_outs\n+ pred_scores, pred_bboxes, anchors, \\\n+ num_anchors_list, stride_tensor_list = head_outs\ngt_labels = gt_meta['gt_class']\ngt_bboxes = gt_meta['gt_bbox']\n+ pad_gt_mask = gt_meta['pad_gt_mask']\n# label assignment\nif gt_meta['epoch_id'] < self.static_assigner_epoch:\nassigned_labels, assigned_bboxes, assigned_scores = self.static_assigner(\n@@ -296,6 +298,7 @@ class TOODHead(nn.Layer):\nnum_anchors_list,\ngt_labels,\ngt_bboxes,\n+ pad_gt_mask,\nbg_index=self.num_classes)\nalpha_l = 0.25\nelse:\n@@ -303,8 +306,10 @@ class TOODHead(nn.Layer):\npred_scores.detach(),\npred_bboxes.detach() * stride_tensor_list,\nbbox_center(anchors),\n+ num_anchors_list,\ngt_labels,\ngt_bboxes,\n+ pad_gt_mask,\nbg_index=self.num_classes)\nalpha_l = -1\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] update assigner and tood_head (#5169)
|
499,339 |
14.02.2022 19:06:44
| -28,800 |
c0b10e6e7d6865c6bc3236b4aeb58935c5b82f34
|
[TIPC] fix benchmark scripts
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/benchmark_train.sh",
"new_path": "test_tipc/benchmark_train.sh",
"diff": "@@ -91,6 +91,7 @@ profile_option=\"${profile_option_key}:${profile_option_params}\"\nline_num=`expr $line_num + 1`\nflags_value=$(func_parser_value \"${lines[line_num]}\")\n+if [ ${flags_value} != \"null\" ];then\n# set flags\nIFS=\";\"\nflags_list=(${flags_value})\n@@ -98,6 +99,7 @@ for _flag in ${flags_list[*]}; do\ncmd=\"export ${_flag}\"\neval $cmd\ndone\n+fi\n# set log_name\nrepo_name=$(get_repo_name )\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/deformable_detr/deformable_detr_r50_1x_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/deformable_detr/deformable_detr_r50_1x_coco_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:2\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:2|8\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/fcos/fcos_r50_fpn_1x_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/fcos/fcos_r50_fpn_1x_coco_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:2|8\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/gfl/gfl_r50_fpn_1x_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/gfl/gfl_r50_fpn_1x_coco_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:2|8\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/keypoint/higherhrnet_hrnet_w32_512_train_infer_python.txt",
"new_path": "test_tipc/configs/keypoint/higherhrnet_hrnet_w32_512_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:20|24\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/keypoint/hrnet_w32_256x192_train_infer_python.txt",
"new_path": "test_tipc/configs/keypoint/hrnet_w32_256x192_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:64|160\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/mot/fairmot_dla34_30e_1088x608_train_infer_python.txt",
"new_path": "test_tipc/configs/mot/fairmot_dla34_30e_1088x608_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:6|22\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/mot/jde_darknet53_30e_1088x608_train_infer_python.txt",
"new_path": "test_tipc/configs/mot/jde_darknet53_30e_1088x608_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:4|14\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/solov2/solov2_r50_fpn_1x_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/solov2/solov2_r50_fpn_1x_coco_train_infer_python.txt",
"diff": "@@ -54,4 +54,4 @@ batch_size:2|4\nfp_items:fp32|fp16\nepoch:1\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n-flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096\n\\ No newline at end of file\n+flags:null\n\\ No newline at end of file\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix benchmark scripts (#5207)
|
499,299 |
17.02.2022 14:41:10
| -28,800 |
9287258124f63a5fdae58e7227de2e379ba31e83
|
fix the order of calculation of center coordinary in ToHeatmapsTopDown
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/transform/keypoint_operators.py",
"new_path": "ppdet/data/transform/keypoint_operators.py",
"diff": "@@ -698,8 +698,8 @@ class ToHeatmapsTopDown(object):\ntmp_size = self.sigma * 3\nfeat_stride = image_size / self.hmsize\nfor joint_id in range(num_joints):\n- mu_x = int(joints[joint_id][0] + 0.5) / feat_stride[0]\n- mu_y = int(joints[joint_id][1] + 0.5) / feat_stride[1]\n+ mu_x = int(joints[joint_id][0] / feat_stride[0] + 0.5)\n+ mu_y = int(joints[joint_id][1] / feat_stride[1] + 0.5)\n# Check that any part of the gaussian is in-bounds\nul = [int(mu_x - tmp_size), int(mu_y - tmp_size)]\nbr = [int(mu_x + tmp_size + 1), int(mu_y + tmp_size + 1)]\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix the order of calculation of center coordinary in ToHeatmapsTopDown (#5222)
|
499,339 |
18.02.2022 20:32:43
| -28,800 |
7b60e7d892247c8bca9d2d12b5e2d8e3bec982f8
|
[TIPC] add benchmark for yolov3, mask_rcnn
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_1x_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_1x_coco_train_infer_python.txt",
"diff": "@@ -49,3 +49,9 @@ inference:./deploy/python/infer.py\n--save_log_path:null\n--run_benchmark:True\n--trt_max_shape:1600\n+===========================train_benchmark_params==========================\n+batch_size:2|4\n+fp_items:fp32|fp16\n+epoch:1\n+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/mask_rcnn/mask_rcnn_r50_fpn_1x_coco_train_infer_python.txt",
"diff": "@@ -49,3 +49,9 @@ inference:./deploy/python/infer.py\n--save_log_path:null\n--run_benchmark:True\n--trt_max_shape:1600\n+===========================train_benchmark_params==========================\n+batch_size:2|4\n+fp_items:fp32|fp16\n+epoch:1\n+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/yolov3/yolov3_darknet53_270e_coco_train_infer_python.txt",
"new_path": "test_tipc/configs/yolov3/yolov3_darknet53_270e_coco_train_infer_python.txt",
"diff": "@@ -49,3 +49,9 @@ inference:./deploy/python/infer.py\n--save_log_path:null\n--run_benchmark:True\nnull:null\n+===========================train_benchmark_params==========================\n+batch_size:8\n+fp_items:fp32|fp16\n+epoch:1\n+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\n+flags:null\n\\ No newline at end of file\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/static/mask_rcnn_r50_1x_coco/N1C1/mask_rcnn_r50_1x_coco_bs2_fp32_SingleP_DP.sh",
"diff": "+model_item=mask_rcnn_r50_1x_coco\n+bs_item=2\n+fp_item=fp32\n+run_process_type=SingleP\n+run_mode=DP\n+device_num=N1C1\n+max_iter=500\n+num_workers=2\n+\n+# get data\n+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh\n+# run\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n+# run profiling\n+sleep 10;\n+export PROFILING=true\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/static/mask_rcnn_r50_1x_coco/N1C8/mask_rcnn_r50_1x_coco_bs2_fp32_MultiP_DP.sh",
"diff": "+model_item=mask_rcnn_r50_1x_coco\n+bs_item=2\n+fp_item=fp32\n+run_process_type=MultiP\n+run_mode=DP\n+device_num=N1C8\n+max_iter=500\n+num_workers=2\n+\n+# get data\n+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh\n+# run\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/static/mask_rcnn_r50_fpn_1x_coco/N1C1/mask_rcnn_r50_fpn_1x_coco_bs2_fp32_SingleP_DP.sh",
"diff": "+model_item=mask_rcnn_r50_fpn_1x_coco\n+bs_item=2\n+fp_item=fp32\n+run_process_type=SingleP\n+run_mode=DP\n+device_num=N1C1\n+max_iter=500\n+num_workers=2\n+\n+# get data\n+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh\n+# run\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n+# run profiling\n+sleep 10;\n+export PROFILING=true\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/static/mask_rcnn_r50_fpn_1x_coco/N1C8/mask_rcnn_r50_fpn_1x_coco_bs2_fp32_MultiP_DP.sh",
"diff": "+model_item=mask_rcnn_r50_fpn_1x_coco\n+bs_item=2\n+fp_item=fp32\n+run_process_type=MultiP\n+run_mode=DP\n+device_num=N1C8\n+max_iter=500\n+num_workers=2\n+\n+# get data\n+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh\n+# run\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/static/yolov3_darknet53_270e_coco/N1C1/yolov3_darknet53_270e_coco_bs8_fp32_SingleP_DP.sh",
"diff": "+model_item=yolov3_darknet53_270e_coco\n+bs_item=8\n+fp_item=fp32\n+run_process_type=SingleP\n+run_mode=DP\n+device_num=N1C1\n+max_iter=500\n+num_workers=8\n+\n+# get data\n+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh\n+# run\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n+# run profiling\n+sleep 10;\n+export PROFILING=true\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/static/yolov3_darknet53_270e_coco/N1C8/yolov3_darknet53_270e_coco_bs8_fp32_MultiP_DP.sh",
"diff": "+model_item=yolov3_darknet53_270e_coco\n+bs_item=8\n+fp_item=fp32\n+run_process_type=MultiP\n+run_mode=DP\n+device_num=N1C8\n+max_iter=500\n+num_workers=8\n+\n+# get data\n+bash test_tipc/static/${model_item}/benchmark_common/prepare.sh\n+# run\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] add benchmark for yolov3, mask_rcnn (#5224)
|
499,352 |
22.02.2022 19:58:53
| -28,800 |
7762b906a5c673a3081818a7d60f717b88ebf202
|
fix blazeface nas
|
[
{
"change_type": "MODIFY",
"old_path": "static/slim/nas/blazeface.yml",
"new_path": "static/slim/nas/blazeface.yml",
"diff": "@@ -93,3 +93,20 @@ EvalReader:\nmean: [104, 117, 123]\nstd: [127.502231, 127.502231, 127.502231]\nbatch_size: 1\n+\n+TestReader:\n+ inputs_def:\n+ fields: ['image', 'im_id', 'im_shape']\n+ dataset:\n+ !ImageFolder\n+ use_default_label: true\n+ sample_transforms:\n+ - !DecodeImage\n+ to_rgb: true\n+ - !NormalizeImage\n+ is_channel_first: false\n+ is_scale: false\n+ mean: [123, 117, 104]\n+ std: [127.502231, 127.502231, 127.502231]\n+ - !Permute {}\n+ batch_size: 1\n"
},
{
"change_type": "MODIFY",
"old_path": "static/slim/nas/search_space/blazefacespace_nas.py",
"new_path": "static/slim/nas/search_space/blazefacespace_nas.py",
"diff": "@@ -33,7 +33,9 @@ class BlazeFaceNasSpace(SearchSpaceBase):\nself.mid_filter_num = np.array([8, 12, 16, 20, 24, 32])\nself.double_filter_num = np.array(\n[8, 12, 16, 24, 32, 40, 48, 64, 72, 80, 88, 96])\n- self.use_5x5kernel = np.array([0, 1])\n+ self.use_5x5kernel = np.array(\n+ [0]\n+ ) ### if constraint is latency, use 3x3 kernel, otherwise self.use_5x5kernel = np.array([0, 1])\ndef init_tokens(self):\nreturn [2, 1, 3, 8, 2, 1, 2, 1, 1]\n@@ -74,7 +76,8 @@ class BlazeFaceNasSpace(SearchSpaceBase):\nself.double_filter_num[tokens[3]]\n]]\n- is_5x5kernel = True if self.use_5x5kernel[tokens[8]] else False\n+ ### if constraint is latency, use 3x3 kernel, otherwise is_5x5kernel = True if self.use_5x5kernel[tokens[8]] else False\n+ is_5x5kernel = False ###True if self.use_5x5kernel[tokens[8]] else False\nreturn blaze_filters, double_blaze_filters, is_5x5kernel\ndef token2arch(self, tokens=None):\n"
},
{
"change_type": "MODIFY",
"old_path": "static/slim/nas/train_nas.py",
"new_path": "static/slim/nas/train_nas.py",
"diff": "@@ -62,6 +62,8 @@ except ImportError as e:\nfrom paddleslim.analysis import flops, TableLatencyEvaluator\nfrom paddleslim.nas import SANAS\n+### register search space to paddleslim\n+import search_space\n@register\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix blazeface nas (#5241)
|
499,301 |
01.03.2022 15:14:41
| -28,800 |
89bfcdaffaaefe7cd24241952b3fc0bc206fb38f
|
rename fp16 -> amp
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -362,8 +362,8 @@ class Trainer(object):\nmodel = paddle.DataParallel(\nself.model, find_unused_parameters=find_unused_parameters)\n- # initial fp16\n- if self.cfg.get('fp16', False):\n+ # enabel auto mixed precision mode\n+ if self.cfg.get('amp', False):\nscaler = amp.GradScaler(\nenable=self.cfg.use_gpu, init_loss_scaling=1024)\n@@ -401,7 +401,7 @@ class Trainer(object):\nself._compose_callback.on_step_begin(self.status)\ndata['epoch_id'] = epoch_id\n- if self.cfg.get('fp16', False):\n+ if self.cfg.get('amp', False):\nwith amp.auto_cast(enable=self.cfg.use_gpu):\n# model forward\noutputs = model(data)\n"
},
{
"change_type": "MODIFY",
"old_path": "tools/train.py",
"new_path": "tools/train.py",
"diff": "@@ -60,10 +60,10 @@ def parse_args():\nhelp=\"If set True, enable continuous evaluation job.\"\n\"This flag is only used for internal test.\")\nparser.add_argument(\n- \"--fp16\",\n+ \"--amp\",\naction='store_true',\ndefault=False,\n- help=\"Enable mixed precision training.\")\n+ help=\"Enable auto mixed precision training.\")\nparser.add_argument(\n\"--fleet\", action='store_true', default=False, help=\"Use fleet or not\")\nparser.add_argument(\n@@ -130,7 +130,7 @@ def run(FLAGS, cfg):\ndef main():\nFLAGS = parse_args()\ncfg = load_config(FLAGS.config)\n- cfg['fp16'] = FLAGS.fp16\n+ cfg['amp'] = FLAGS.amp\ncfg['fleet'] = FLAGS.fleet\ncfg['use_vdl'] = FLAGS.use_vdl\ncfg['vdl_log_dir'] = FLAGS.vdl_log_dir\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
rename fp16 -> amp (#5268)
|
499,298 |
01.03.2022 15:16:00
| -28,800 |
629e1533b9a9bfb58df9e4595a36539b22afaa63
|
fix doc deadlinks
|
[
{
"change_type": "RENAME",
"old_path": "configs/centernet/centernet_mbv1_1x_140e_coco.yml",
"new_path": "configs/centernet/centernet_mbv1_140e_coco.yml",
"diff": "@@ -3,7 +3,7 @@ _BASE_: [\n]\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/MobileNetV1_pretrained.pdparams\n-weights: output/centernet_mbv1_1x_140e_coco/model_final\n+weights: output/centernet_mbv1_140e_coco/model_final\nCenterNet:\nbackbone: MobileNet\n"
},
{
"change_type": "RENAME",
"old_path": "configs/centernet/centernet_mbv3_large_1x_140e_coco.yml",
"new_path": "configs/centernet/centernet_mbv3_large_140e_coco.yml",
"diff": "@@ -3,7 +3,7 @@ _BASE_: [\n]\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/MobileNetV3_large_x1_0_ssld_pretrained.pdparams\n-weights: output/centernet_mbv3_large_1x_140e_coco/model_final\n+weights: output/centernet_mbv3_large_140e_coco/model_final\nCenterNet:\nbackbone: MobileNetV3\n"
},
{
"change_type": "RENAME",
"old_path": "configs/centernet/centernet_mbv3_small_1x_140e_coco.yml",
"new_path": "configs/centernet/centernet_mbv3_small_140e_coco.yml",
"diff": "@@ -3,7 +3,7 @@ _BASE_: [\n]\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/MobileNetV3_small_x1_0_ssld_pretrained.pdparams\n-weights: output/centernet_mbv3_small_1x_140e_coco/model_final\n+weights: output/centernet_mbv3_small_140e_coco/model_final\nCenterNet:\nbackbone: MobileNetV3\n@@ -24,6 +24,5 @@ CenterNetDLAFPN:\ndown_ratio: 8\ndcn_v2: False\n-\nTrainReader:\nbatch_size: 32\n"
},
{
"change_type": "RENAME",
"old_path": "configs/centernet/centernet_shufflenetv2_1x_140e_coco.yml",
"new_path": "configs/centernet/centernet_shufflenetv2_140e_coco.yml",
"diff": "@@ -3,7 +3,7 @@ _BASE_: [\n]\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ShuffleNetV2_x1_0_pretrained.pdparams\n-weights: output/centernet_shufflenetv2_1x_140e_coco/model_final\n+weights: output/centernet_shufflenetv2_140e_coco/model_final\nCenterNet:\nbackbone: ShuffleNetV2\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix doc deadlinks (#5270)
|
499,339 |
03.03.2022 18:36:04
| -28,800 |
e9cfb1876e5dabfd116526639911905cc8e7a6fd
|
[DETR] fix find_unused_parameters in yaml
|
[
{
"change_type": "MODIFY",
"old_path": "configs/deformable_detr/README.md",
"new_path": "configs/deformable_detr/README.md",
"diff": "@@ -20,7 +20,7 @@ Deformable DETR is an object detection model based on DETR. We reproduced the mo\nGPU multi-card training\n```bash\nexport CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7\n-python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/deformable_detr/deformable_detr_r50_1x_coco.yml --fleet -o find_unused_parameters=True\n+python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/deformable_detr/deformable_detr_r50_1x_coco.yml --fleet\n```\n## Citations\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/deformable_detr/deformable_detr_r50_1x_coco.yml",
"new_path": "configs/deformable_detr/deformable_detr_r50_1x_coco.yml",
"diff": "@@ -6,3 +6,4 @@ _BASE_: [\n'_base_/deformable_detr_reader.yml',\n]\nweights: output/deformable_detr_r50_1x_coco/model_final\n+find_unused_parameters: True\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/detr/README.md",
"new_path": "configs/detr/README.md",
"diff": "@@ -20,7 +20,7 @@ DETR is an object detection model based on transformer. We reproduced the model\nGPU multi-card training\n```bash\nexport CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7\n-python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/detr/detr_r50_1x_coco.yml --fleet -o find_unused_parameters=True\n+python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/detr/detr_r50_1x_coco.yml --fleet\n```\n## Citations\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/detr/detr_r50_1x_coco.yml",
"new_path": "configs/detr/detr_r50_1x_coco.yml",
"diff": "@@ -6,3 +6,4 @@ _BASE_: [\n'_base_/detr_reader.yml',\n]\nweights: output/detr_r50_1x_coco/model_final\n+find_unused_parameters: True\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[DETR] fix find_unused_parameters in yaml (#5291)
|
499,304 |
04.03.2022 09:49:49
| -28,800 |
faa4f9a8e71bdb94ad3c6415dbefe0abd98a27ea
|
fix picodet post_process
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/gfl_head.py",
"new_path": "ppdet/modeling/heads/gfl_head.py",
"diff": "@@ -430,9 +430,7 @@ class GFLHead(nn.Layer):\ncls_scores, bboxes_reg = gfl_head_outs\nbboxes = paddle.concat(bboxes_reg, axis=1)\n# rescale: [h_scale, w_scale] -> [w_scale, h_scale, w_scale, h_scale]\n- im_scale = paddle.concat(\n- [scale_factor[:, ::-1], scale_factor[:, ::-1]],\n- axis=-1).unsqueeze(1)\n+ im_scale = scale_factor.flip([1]).tile([1, 2]).unsqueeze(1)\nbboxes /= im_scale\nmlvl_scores = paddle.concat(cls_scores, axis=1)\nmlvl_scores = mlvl_scores.transpose([0, 2, 1])\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix picodet post_process (#5292)
|
499,339 |
04.03.2022 11:15:42
| -28,800 |
161d5bbec5b8430afdeffdae716fd5dba3d7a251
|
[TIPC] fix benchmark static shell, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/static/mask_rcnn_r50_1x_coco/N1C1/mask_rcnn_r50_1x_coco_bs2_fp32_SingleP_DP.sh",
"new_path": "test_tipc/static/mask_rcnn_r50_1x_coco/N1C1/mask_rcnn_r50_1x_coco_bs2_fp32_SingleP_DP.sh",
"diff": "@@ -4,7 +4,7 @@ fp_item=fp32\nrun_process_type=SingleP\nrun_mode=DP\ndevice_num=N1C1\n-max_iter=500\n+max_iter=100\nnum_workers=2\n# get data\n@@ -14,4 +14,4 @@ bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_it\n# run profiling\nsleep 10;\nexport PROFILING=true\n-bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} 11 ${num_workers} 2>&1;\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/static/mask_rcnn_r50_1x_coco/N1C8/mask_rcnn_r50_1x_coco_bs2_fp32_MultiP_DP.sh",
"new_path": "test_tipc/static/mask_rcnn_r50_1x_coco/N1C8/mask_rcnn_r50_1x_coco_bs2_fp32_MultiP_DP.sh",
"diff": "@@ -4,7 +4,7 @@ fp_item=fp32\nrun_process_type=MultiP\nrun_mode=DP\ndevice_num=N1C8\n-max_iter=500\n+max_iter=100\nnum_workers=2\n# get data\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/static/mask_rcnn_r50_fpn_1x_coco/N1C1/mask_rcnn_r50_fpn_1x_coco_bs2_fp32_SingleP_DP.sh",
"new_path": "test_tipc/static/mask_rcnn_r50_fpn_1x_coco/N1C1/mask_rcnn_r50_fpn_1x_coco_bs2_fp32_SingleP_DP.sh",
"diff": "@@ -4,7 +4,7 @@ fp_item=fp32\nrun_process_type=SingleP\nrun_mode=DP\ndevice_num=N1C1\n-max_iter=500\n+max_iter=100\nnum_workers=2\n# get data\n@@ -14,4 +14,4 @@ bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_it\n# run profiling\nsleep 10;\nexport PROFILING=true\n-bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} 11 ${num_workers} 2>&1;\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/static/mask_rcnn_r50_fpn_1x_coco/N1C8/mask_rcnn_r50_fpn_1x_coco_bs2_fp32_MultiP_DP.sh",
"new_path": "test_tipc/static/mask_rcnn_r50_fpn_1x_coco/N1C8/mask_rcnn_r50_fpn_1x_coco_bs2_fp32_MultiP_DP.sh",
"diff": "@@ -4,7 +4,7 @@ fp_item=fp32\nrun_process_type=MultiP\nrun_mode=DP\ndevice_num=N1C8\n-max_iter=500\n+max_iter=100\nnum_workers=2\n# get data\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/static/yolov3_darknet53_270e_coco/N1C1/yolov3_darknet53_270e_coco_bs8_fp32_SingleP_DP.sh",
"new_path": "test_tipc/static/yolov3_darknet53_270e_coco/N1C1/yolov3_darknet53_270e_coco_bs8_fp32_SingleP_DP.sh",
"diff": "@@ -4,7 +4,7 @@ fp_item=fp32\nrun_process_type=SingleP\nrun_mode=DP\ndevice_num=N1C1\n-max_iter=500\n+max_iter=100\nnum_workers=8\n# get data\n@@ -14,4 +14,4 @@ bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_it\n# run profiling\nsleep 10;\nexport PROFILING=true\n-bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} ${max_iter} ${num_workers} 2>&1;\n+bash test_tipc/static/${model_item}/benchmark_common/run_benchmark.sh ${model_item} ${bs_item} ${fp_item} ${run_process_type} ${run_mode} ${device_num} 11 ${num_workers} 2>&1;\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/static/yolov3_darknet53_270e_coco/N1C8/yolov3_darknet53_270e_coco_bs8_fp32_MultiP_DP.sh",
"new_path": "test_tipc/static/yolov3_darknet53_270e_coco/N1C8/yolov3_darknet53_270e_coco_bs8_fp32_MultiP_DP.sh",
"diff": "@@ -4,7 +4,7 @@ fp_item=fp32\nrun_process_type=MultiP\nrun_mode=DP\ndevice_num=N1C8\n-max_iter=500\n+max_iter=100\nnum_workers=8\n# get data\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix benchmark static shell, test=document_fix (#5293)
|
499,299 |
05.03.2022 21:49:22
| -28,800 |
fca31c8133ab2c13504256bbb2d843050c9cb3e8
|
fix python inference error when use fuse_norm keypoint model
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/keypoint_infer.py",
"new_path": "deploy/python/keypoint_infer.py",
"diff": "@@ -288,7 +288,7 @@ def create_inputs(imgs, im_info):\ninputs (dict): input of model\n\"\"\"\ninputs = {}\n- inputs['image'] = np.stack(imgs, axis=0)\n+ inputs['image'] = np.stack(imgs, axis=0).astype('float32')\nim_shape = []\nfor e in im_info:\nim_shape.append(np.array((e['im_shape'])).astype('float32'))\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix python inference error when use fuse_norm keypoint model (#5300)
|
499,299 |
05.03.2022 21:49:33
| -28,800 |
a4e093b799af582df74079584f1fa3ae07c1c538
|
fix use_dark can not set to False
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/det_keypoint_unite_utils.py",
"new_path": "deploy/python/det_keypoint_unite_utils.py",
"diff": "@@ -112,7 +112,7 @@ def argsparser():\n\"calibration, trt_calib_mode need to set True.\")\nparser.add_argument(\n'--use_dark',\n- type=bool,\n+ type=ast.literal_eval,\ndefault=True,\nhelp='whether to use darkpose to get better keypoint position predict ')\nparser.add_argument(\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/utils.py",
"new_path": "deploy/python/utils.py",
"diff": "@@ -133,7 +133,7 @@ def argsparser():\nhelp=\"max batch_size for reid model inference.\")\nparser.add_argument(\n'--use_dark',\n- type=bool,\n+ type=ast.literal_eval,\ndefault=True,\nhelp='whether to use darkpose to get better keypoint position predict ')\nreturn parser\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix use_dark can not set to False (#5304)
|
499,304 |
07.03.2022 10:48:35
| -28,800 |
c27233d313fa7ddf3da73725eff71fac65f8d8e4
|
fix picodetv2 post process
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/pico_head.py",
"new_path": "ppdet/modeling/heads/pico_head.py",
"diff": "@@ -460,18 +460,15 @@ class PicoHeadV2(GFLHead):\nact=self.act,\nuse_act_in_out=False))\n- def forward(self, fpn_feats, deploy=False):\n+ def forward(self, fpn_feats, export_post_process=True):\nassert len(fpn_feats) == len(\nself.fpn_stride\n), \"The size of fpn_feats is not equal to size of fpn_stride\"\n- anchors, _, num_anchors_list, stride_tensor_list = generate_anchors_for_grid_cell(\n- fpn_feats, self.fpn_stride, self.grid_cell_scale, self.cell_offset)\n- anchors_split = paddle.split(anchors, num_anchors_list)\ncls_score_list, reg_list, box_list = [], [], []\n- for i, fpn_feat, anchor, stride, align_cls in zip(\n- range(len(self.fpn_stride)), fpn_feats, anchors_split,\n- self.fpn_stride, self.cls_align):\n+ for i, fpn_feat, stride, align_cls in zip(\n+ range(len(self.fpn_stride)), fpn_feats, self.fpn_stride,\n+ self.cls_align):\nb, _, h, w = get_static_shape(fpn_feat)\n# task decomposition\nconv_cls_feat, se_feat = self.conv_feat(fpn_feat, i)\n@@ -485,22 +482,35 @@ class PicoHeadV2(GFLHead):\nelse:\ncls_score = F.sigmoid(cls_logit)\n- anchor_centers = bbox_center(anchor).unsqueeze(0) / stride\n- anchor_centers = anchor_centers.reshape([1, h, w, 2])\n-\n- pred_distances = self.distribution_project(\n- reg_pred.transpose([0, 2, 3, 1])).reshape([b, h, w, 4])\n- reg_bbox = batch_distance2bbox(\n- anchor_centers, pred_distances, max_shapes=None)\n- if not self.training:\n+ if not export_post_process and not self.training:\n+ # Now only supports batch size = 1 in deploy\ncls_score_list.append(\n- cls_score.transpose([0, 2, 3, 1]).reshape(\n- [b, -1, self.cls_out_channels]))\n- box_list.append(reg_bbox.reshape([b, -1, 4]) * stride)\n+ cls_score.reshape([1, self.cls_out_channels, -1]).transpose(\n+ [0, 2, 1]))\n+ box_list.append(\n+ reg_pred.reshape([1, (self.reg_max + 1) * 4, -1]).transpose(\n+ [0, 2, 1]))\nelse:\n- cls_score_list.append(cls_score.flatten(2).transpose([0, 2, 1]))\n+ cls_score_out = cls_score.transpose([0, 2, 3, 1])\n+ bbox_pred = reg_pred.transpose([0, 2, 3, 1])\n+ b, cell_h, cell_w, _ = paddle.shape(cls_score_out)\n+ y, x = self.get_single_level_center_point(\n+ [cell_h, cell_w], stride, cell_offset=self.cell_offset)\n+ center_points = paddle.stack([x, y], axis=-1)\n+ cls_score_out = cls_score_out.reshape(\n+ [b, -1, self.cls_out_channels])\n+ bbox_pred = self.distribution_project(bbox_pred) * stride\n+ bbox_pred = bbox_pred.reshape([b, cell_h * cell_w, 4])\n+ bbox_pred = batch_distance2bbox(\n+ center_points, bbox_pred, max_shapes=None)\n+ if not self.training:\n+ cls_score_list.append(cls_score_out)\n+ box_list.append(bbox_pred)\n+ else:\n+ cls_score_list.append(\n+ cls_score.flatten(2).transpose([0, 2, 1]))\nreg_list.append(reg_pred.flatten(2).transpose([0, 2, 1]))\n- box_list.append(reg_bbox.reshape([b, -1, 4]))\n+ box_list.append(bbox_pred / stride)\nif not self.training:\nreturn cls_score_list, box_list\n@@ -508,16 +518,19 @@ class PicoHeadV2(GFLHead):\ncls_score_list = paddle.concat(cls_score_list, axis=1)\nbox_list = paddle.concat(box_list, axis=1)\nreg_list = paddle.concat(reg_list, axis=1)\n- return cls_score_list, reg_list, box_list, anchors, num_anchors_list, stride_tensor_list\n+ return cls_score_list, reg_list, box_list, fpn_feats\ndef get_loss(self, head_outs, gt_meta):\n- pred_scores, pred_regs, pred_bboxes, anchors, num_anchors_list, stride_tensor_list = head_outs\n+ pred_scores, pred_regs, pred_bboxes, fpn_feats = head_outs\ngt_labels = gt_meta['gt_class']\ngt_bboxes = gt_meta['gt_bbox']\ngt_scores = gt_meta['gt_score'] if 'gt_score' in gt_meta else None\nnum_imgs = gt_meta['im_id'].shape[0]\npad_gt_mask = gt_meta['pad_gt_mask']\n+ anchors, _, num_anchors_list, stride_tensor_list = generate_anchors_for_grid_cell(\n+ fpn_feats, self.fpn_stride, self.grid_cell_scale, self.cell_offset)\n+\ncenters = bbox_center(anchors)\n# label assignment\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix picodetv2 post process (#5306)
|
499,339 |
07.03.2022 17:12:54
| -28,800 |
34a74b39dc439d1f13925ca70a9418a1fee477fa
|
refine ema_model save
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/callbacks.py",
"new_path": "ppdet/engine/callbacks.py",
"diff": "@@ -182,7 +182,7 @@ class Checkpointer(Callback):\n) % self.model.cfg.snapshot_epoch == 0 or epoch_id == end_epoch - 1:\nsave_name = str(\nepoch_id) if epoch_id != end_epoch - 1 else \"model_final\"\n- weight = self.weight\n+ weight = self.weight.state_dict()\nelif mode == 'eval':\nif 'save_best_model' in status and status['save_best_model']:\nfor metric in self.model._metrics:\n@@ -201,18 +201,22 @@ class Checkpointer(Callback):\nif map_res[key][0] > self.best_ap:\nself.best_ap = map_res[key][0]\nsave_name = 'best_model'\n- weight = self.weight\n+ weight = self.weight.state_dict()\nlogger.info(\"Best test {} ap is {:0.3f}.\".format(\nkey, self.best_ap))\nif weight:\nif self.model.use_ema:\n- save_model(status['weight'], self.save_dir, save_name,\n- epoch_id + 1, self.model.optimizer)\n- save_model(weight, self.save_dir,\n- '{}_ema'.format(save_name), epoch_id + 1)\n+ # save model and ema_model\n+ save_model(\n+ status['weight'],\n+ self.model.optimizer,\n+ self.save_dir,\n+ save_name,\n+ epoch_id + 1,\n+ ema_model=weight)\nelse:\n- save_model(weight, self.save_dir, save_name, epoch_id + 1,\n- self.model.optimizer)\n+ save_model(weight, self.model.optimizer, self.save_dir,\n+ save_name, epoch_id + 1)\nclass WiferFaceEval(Callback):\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer.py",
"diff": "@@ -332,7 +332,7 @@ class ModelEMA(object):\nfor k, v in self.state_dict.items():\nself.state_dict[k] = paddle.zeros_like(v)\n- def resume(self, state_dict, step):\n+ def resume(self, state_dict, step=0):\nfor k, v in state_dict.items():\nself.state_dict[k] = v\nself.step = step\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/utils/checkpoint.py",
"new_path": "ppdet/utils/checkpoint.py",
"diff": "@@ -72,7 +72,14 @@ def load_weight(model, weight, optimizer=None, ema=None):\nraise ValueError(\"Model pretrain path {} does not \"\n\"exists.\".format(pdparam_path))\n+ if ema is not None and os.path.exists(path + '.pdema'):\n+ # Exchange model and ema_model to load\n+ ema_state_dict = paddle.load(pdparam_path)\n+ param_state_dict = paddle.load(path + '.pdema')\n+ else:\n+ ema_state_dict = None\nparam_state_dict = paddle.load(pdparam_path)\n+\nmodel_dict = model.state_dict()\nmodel_weight = {}\nincorrect_keys = 0\n@@ -102,10 +109,11 @@ def load_weight(model, weight, optimizer=None, ema=None):\nlast_epoch = optim_state_dict.pop('last_epoch')\noptimizer.set_state_dict(optim_state_dict)\n- if ema is not None and os.path.exists(path + '_ema.pdparams'):\n- ema_state_dict = paddle.load(path + '_ema.pdparams')\n+ if ema_state_dict is not None:\nema.resume(ema_state_dict,\noptim_state_dict['LR_Scheduler']['last_epoch'])\n+ elif ema_state_dict is not None:\n+ ema.resume(ema_state_dict)\nreturn last_epoch\n@@ -205,30 +213,42 @@ def load_pretrain_weight(model, pretrain_weight):\nlogger.info('Finish loading model weights: {}'.format(weights_path))\n-def save_model(model, save_dir, save_name, last_epoch, optimizer=None):\n+def save_model(model,\n+ optimizer,\n+ save_dir,\n+ save_name,\n+ last_epoch,\n+ ema_model=None):\n\"\"\"\nsave model into disk.\nArgs:\n- model (paddle.nn.Layer): the Layer instalce to save parameters.\n+ model (dict): the model state_dict to save parameters.\noptimizer (paddle.optimizer.Optimizer): the Optimizer instance to\nsave optimizer states.\nsave_dir (str): the directory to be saved.\nsave_name (str): the path to be saved.\nlast_epoch (int): the epoch index.\n+ ema_model (dict|None): the ema_model state_dict to save parameters.\n\"\"\"\nif paddle.distributed.get_rank() != 0:\nreturn\n+ assert isinstance(model, dict), (\"model is not a instance of dict, \"\n+ \"please call model.state_dict() to get.\")\nif not os.path.exists(save_dir):\nos.makedirs(save_dir)\nsave_path = os.path.join(save_dir, save_name)\n- if isinstance(model, nn.Layer):\n- paddle.save(model.state_dict(), save_path + \".pdparams\")\n- else:\n- assert isinstance(model,\n- dict), 'model is not a instance of nn.layer or dict'\n+ # save model\n+ if ema_model is None:\npaddle.save(model, save_path + \".pdparams\")\n- if optimizer is not None:\n+ else:\n+ assert isinstance(ema_model,\n+ dict), (\"ema_model is not a instance of dict, \"\n+ \"please call model.state_dict() to get.\")\n+ # Exchange model and ema_model to save\n+ paddle.save(ema_model, save_path + \".pdparams\")\n+ paddle.save(model, save_path + \".pdema\")\n+ # save optimizer\nstate_dict = optimizer.state_dict()\nstate_dict['last_epoch'] = last_epoch\npaddle.save(state_dict, save_path + \".pdopt\")\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
refine ema_model save (#5314)
|
499,298 |
08.03.2022 12:03:05
| -28,800 |
56d22694f75b00e89cf694ccc069e4a027b2b6fb
|
[MOT] unify mot and det output format
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot/tracker/deepsort_tracker.py",
"new_path": "deploy/pptracking/python/mot/tracker/deepsort_tracker.py",
"diff": "@@ -90,13 +90,13 @@ class DeepSORTTracker(object):\nPerform measurement update and track management.\nArgs:\npred_dets (np.array): Detection results of the image, the shape is\n- [N, 6], means 'x0, y0, x1, y1, score, cls_id'.\n+ [N, 6], means 'cls_id, score, x0, y0, x1, y1'.\npred_embs (np.array): Embedding results of the image, the shape is\n[N, 128], usually pred_embs.shape[1] is a multiple of 128.\n\"\"\"\n- pred_tlwhs = pred_dets[:, :4]\n- pred_scores = pred_dets[:, 4:5]\n- pred_cls_ids = pred_dets[:, 5:]\n+ pred_cls_ids = pred_dets[:, 0:1]\n+ pred_scores = pred_dets[:, 1:2]\n+ pred_tlwhs = pred_dets[:, 2:6]\ndetections = [\nDetection(tlwh, score, feat, cls_id)\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot/tracker/jde_tracker.py",
"new_path": "deploy/pptracking/python/mot/tracker/jde_tracker.py",
"diff": "@@ -100,7 +100,7 @@ class JDETracker(object):\nArgs:\npred_dets (np.array): Detection results of the image, the shape is\n- [N, 6], means 'x0, y0, x1, y1, score, cls_id'.\n+ [N, 6], means 'cls_id, score, x0, y0, x1, y1'.\npred_embs (np.array): Embedding results of the image, the shape is\n[N, 128] or [N, 512].\n@@ -122,7 +122,7 @@ class JDETracker(object):\n# unify single and multi classes detection and embedding results\nfor cls_id in range(self.num_classes):\n- cls_idx = (pred_dets[:, 5:] == cls_id).squeeze(-1)\n+ cls_idx = (pred_dets[:, 0:1] == cls_id).squeeze(-1)\npred_dets_dict[cls_id] = pred_dets[cls_idx]\nif pred_embs is not None:\npred_embs_dict[cls_id] = pred_embs[cls_idx]\n@@ -133,21 +133,26 @@ class JDETracker(object):\n\"\"\" Step 1: Get detections by class\"\"\"\npred_dets_cls = pred_dets_dict[cls_id]\npred_embs_cls = pred_embs_dict[cls_id]\n- remain_inds = (pred_dets_cls[:, 4:5] > self.conf_thres).squeeze(-1)\n+ remain_inds = (pred_dets_cls[:, 1:2] > self.conf_thres).squeeze(-1)\nif remain_inds.sum() > 0:\npred_dets_cls = pred_dets_cls[remain_inds]\nif self.use_byte:\ndetections = [\nSTrack(\n- STrack.tlbr_to_tlwh(tlbrs[:4]), tlbrs[4], cls_id, 30, temp_feat=None)\n- for tlbrs in pred_dets_cls\n+ STrack.tlbr_to_tlwh(tlbrs[2:6]),\n+ tlbrs[1],\n+ cls_id,\n+ 30,\n+ temp_feat=None) for tlbrs in pred_dets_cls\n]\nelse:\npred_embs_cls = pred_embs_cls[remain_inds]\ndetections = [\nSTrack(\n- STrack.tlbr_to_tlwh(tlbrs[:4]), tlbrs[4], cls_id, 30, temp_feat)\n- for (tlbrs, temp_feat) in zip(pred_dets_cls, pred_embs_cls)\n+ STrack.tlbr_to_tlwh(tlbrs[2:6]), tlbrs[1], cls_id,\n+ 30, temp_feat)\n+ for (tlbrs, temp_feat\n+ ) in zip(pred_dets_cls, pred_embs_cls)\n]\nelse:\ndetections = []\n@@ -171,14 +176,17 @@ class JDETracker(object):\nSTrack.multi_predict(track_pool_dict[cls_id], self.motion)\nif self.use_byte:\n- dists = matching.iou_distance(track_pool_dict[cls_id], detections)\n+ dists = matching.iou_distance(track_pool_dict[cls_id],\n+ detections)\nmatches, u_track, u_detection = matching.linear_assignment(\ndists, thresh=self.match_thres) # not self.tracked_thresh\nelse:\ndists = matching.embedding_distance(\n- track_pool_dict[cls_id], detections, metric=self.metric_type)\n- dists = matching.fuse_motion(self.motion, dists,\n- track_pool_dict[cls_id], detections)\n+ track_pool_dict[cls_id],\n+ detections,\n+ metric=self.metric_type)\n+ dists = matching.fuse_motion(\n+ self.motion, dists, track_pool_dict[cls_id], detections)\nmatches, u_track, u_detection = matching.linear_assignment(\ndists, thresh=self.tracked_thresh)\n@@ -199,15 +207,20 @@ class JDETracker(object):\n# None of the steps below happen if there are no undetected tracks.\n\"\"\" Step 3: Second association, with IOU\"\"\"\nif self.use_byte:\n- inds_low = pred_dets_dict[cls_id][:, 4:5] > self.low_conf_thres\n- inds_high = pred_dets_dict[cls_id][:, 4:5] < self.conf_thres\n+ inds_low = pred_dets_dict[cls_id][:, 1:2] > self.low_conf_thres\n+ inds_high = pred_dets_dict[cls_id][:, 1:2] < self.conf_thres\ninds_second = np.logical_and(inds_low, inds_high).squeeze(-1)\npred_dets_cls_second = pred_dets_dict[cls_id][inds_second]\n# association the untrack to the low score detections\nif len(pred_dets_cls_second) > 0:\ndetections_second = [\n- STrack(STrack.tlbr_to_tlwh(tlbrs[:4]), tlbrs[4], cls_id, 30, temp_feat=None)\n+ STrack(\n+ STrack.tlbr_to_tlwh(tlbrs[:4]),\n+ tlbrs[4],\n+ cls_id,\n+ 30,\n+ temp_feat=None)\nfor tlbrs in pred_dets_cls_second[:, :5]\n]\nelse:\n@@ -216,7 +229,8 @@ class JDETracker(object):\ntrack_pool_dict[cls_id][i] for i in u_track\nif track_pool_dict[cls_id][i].state == TrackState.Tracked\n]\n- dists = matching.iou_distance(r_tracked_stracks, detections_second)\n+ dists = matching.iou_distance(r_tracked_stracks,\n+ detections_second)\nmatches, u_track, u_detection_second = matching.linear_assignment(\ndists, thresh=0.4) # not r_tracked_thresh\nelse:\n@@ -232,7 +246,8 @@ class JDETracker(object):\nfor i_tracked, idet in matches:\ntrack = r_tracked_stracks[i_tracked]\n- det = detections[idet] if not self.use_byte else detections_second[idet]\n+ det = detections[\n+ idet] if not self.use_byte else detections_second[idet]\nif track.state == TrackState.Tracked:\ntrack.update(det, self.frame_id)\nactivated_tracks_dict[cls_id].append(track)\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/mot_jde_infer.py",
"new_path": "deploy/python/mot_jde_infer.py",
"diff": "@@ -115,7 +115,7 @@ class JDE_Detector(Detector):\nreturn result\ndef tracking(self, det_results):\n- pred_dets = det_results['pred_dets']\n+ pred_dets = det_results['pred_dets'] # 'cls_id, score, x0, y0, x1, y1'\npred_embs = det_results['pred_embs']\nonline_targets_dict = self.tracker.update(pred_dets, pred_embs)\n@@ -143,7 +143,7 @@ class JDE_Detector(Detector):\nrepeats (int): repeats number for prediction\nReturns:\nresult (dict): include 'pred_dets': np.ndarray: shape:[N,6], N: number of box,\n- matix element:[x_min, y_min, x_max, y_max, score, class]\n+ matix element:[class, score, x_min, y_min, x_max, y_max]\nFairMOT(JDE)'s result include 'pred_embs': np.ndarray:\nshape: [N, 128]\n'''\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/mot_sde_infer.py",
"new_path": "deploy/python/mot_sde_infer.py",
"diff": "@@ -111,11 +111,8 @@ class SDE_Detector(Detector):\nlow_conf_thres=low_conf_thres)\ndef tracking(self, det_results):\n- pred_dets = det_results['boxes']\n+ pred_dets = det_results['boxes'] # 'cls_id, score, x0, y0, x1, y1'\npred_embs = None\n- pred_dets = np.concatenate(\n- (pred_dets[:, 2:], pred_dets[:, 1:2], pred_dets[:, 0:1]), 1)\n- # pred_dets should be 'x0, y0, x1, y1, score, cls_id'\nonline_targets_dict = self.tracker.update(pred_dets, pred_embs)\nonline_tlwhs = defaultdict(list)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/tracker.py",
"new_path": "ppdet/engine/tracker.py",
"diff": "@@ -282,14 +282,14 @@ class Tracker(object):\n# thus will not inference reid model\ncontinue\n- pred_scores = pred_scores[keep_idx[0]]\npred_cls_ids = pred_cls_ids[keep_idx[0]]\n+ pred_scores = pred_scores[keep_idx[0]]\npred_tlwhs = np.concatenate(\n(pred_xyxys[:, 0:2],\npred_xyxys[:, 2:4] - pred_xyxys[:, 0:2] + 1),\naxis=1)\npred_dets = np.concatenate(\n- (pred_tlwhs, pred_scores, pred_cls_ids), axis=1)\n+ (pred_cls_ids, pred_scores, pred_tlwhs), axis=1)\ntracker = self.model.tracker\ncrops = get_crops(\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/mot/tracker/deepsort_tracker.py",
"new_path": "ppdet/modeling/mot/tracker/deepsort_tracker.py",
"diff": "@@ -96,13 +96,13 @@ class DeepSORTTracker(object):\nPerform measurement update and track management.\nArgs:\npred_dets (np.array): Detection results of the image, the shape is\n- [N, 6], means 'x0, y0, x1, y1, score, cls_id'.\n+ [N, 6], means 'cls_id, score, x0, y0, x1, y1'.\npred_embs (np.array): Embedding results of the image, the shape is\n[N, 128], usually pred_embs.shape[1] is a multiple of 128.\n\"\"\"\n- pred_tlwhs = pred_dets[:, :4]\n- pred_scores = pred_dets[:, 4:5]\n- pred_cls_ids = pred_dets[:, 5:]\n+ pred_cls_ids = pred_dets[:, 0:1]\n+ pred_scores = pred_dets[:, 1:2]\n+ pred_tlwhs = pred_dets[:, 2:6]\ndetections = [\nDetection(tlwh, score, feat, cls_id)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/mot/tracker/jde_tracker.py",
"new_path": "ppdet/modeling/mot/tracker/jde_tracker.py",
"diff": "@@ -106,7 +106,7 @@ class JDETracker(object):\nArgs:\npred_dets (np.array): Detection results of the image, the shape is\n- [N, 6], means 'x0, y0, x1, y1, score, cls_id'.\n+ [N, 6], means 'cls_id, score, x0, y0, x1, y1'.\npred_embs (np.array): Embedding results of the image, the shape is\n[N, 128] or [N, 512].\n@@ -128,7 +128,7 @@ class JDETracker(object):\n# unify single and multi classes detection and embedding results\nfor cls_id in range(self.num_classes):\n- cls_idx = (pred_dets[:, 5:] == cls_id).squeeze(-1)\n+ cls_idx = (pred_dets[:, 0:1] == cls_id).squeeze(-1)\npred_dets_dict[cls_id] = pred_dets[cls_idx]\nif pred_embs is not None:\npred_embs_dict[cls_id] = pred_embs[cls_idx]\n@@ -139,21 +139,26 @@ class JDETracker(object):\n\"\"\" Step 1: Get detections by class\"\"\"\npred_dets_cls = pred_dets_dict[cls_id]\npred_embs_cls = pred_embs_dict[cls_id]\n- remain_inds = (pred_dets_cls[:, 4:5] > self.conf_thres).squeeze(-1)\n+ remain_inds = (pred_dets_cls[:, 1:2] > self.conf_thres).squeeze(-1)\nif remain_inds.sum() > 0:\npred_dets_cls = pred_dets_cls[remain_inds]\nif self.use_byte:\ndetections = [\nSTrack(\n- STrack.tlbr_to_tlwh(tlbrs[:4]), tlbrs[4], cls_id, 30, temp_feat=None)\n- for tlbrs in pred_dets_cls\n+ STrack.tlbr_to_tlwh(tlbrs[2:6]),\n+ tlbrs[1],\n+ cls_id,\n+ 30,\n+ temp_feat=None) for tlbrs in pred_dets_cls\n]\nelse:\npred_embs_cls = pred_embs_cls[remain_inds]\ndetections = [\nSTrack(\n- STrack.tlbr_to_tlwh(tlbrs[:4]), tlbrs[4], cls_id, 30, temp_feat)\n- for (tlbrs, temp_feat) in zip(pred_dets_cls, pred_embs_cls)\n+ STrack.tlbr_to_tlwh(tlbrs[2:6]), tlbrs[1], cls_id,\n+ 30, temp_feat)\n+ for (tlbrs, temp_feat\n+ ) in zip(pred_dets_cls, pred_embs_cls)\n]\nelse:\ndetections = []\n@@ -177,14 +182,17 @@ class JDETracker(object):\nSTrack.multi_predict(track_pool_dict[cls_id], self.motion)\nif self.use_byte:\n- dists = matching.iou_distance(track_pool_dict[cls_id], detections)\n+ dists = matching.iou_distance(track_pool_dict[cls_id],\n+ detections)\nmatches, u_track, u_detection = matching.linear_assignment(\n- dists, thresh=self.match_thres) #\n+ dists, thresh=self.match_thres) # not self.tracked_thresh\nelse:\ndists = matching.embedding_distance(\n- track_pool_dict[cls_id], detections, metric=self.metric_type)\n- dists = matching.fuse_motion(self.motion, dists,\n- track_pool_dict[cls_id], detections)\n+ track_pool_dict[cls_id],\n+ detections,\n+ metric=self.metric_type)\n+ dists = matching.fuse_motion(\n+ self.motion, dists, track_pool_dict[cls_id], detections)\nmatches, u_track, u_detection = matching.linear_assignment(\ndists, thresh=self.tracked_thresh)\n@@ -205,15 +213,20 @@ class JDETracker(object):\n# None of the steps below happen if there are no undetected tracks.\n\"\"\" Step 3: Second association, with IOU\"\"\"\nif self.use_byte:\n- inds_low = pred_dets_dict[cls_id][:, 4:5] > self.low_conf_thres\n- inds_high = pred_dets_dict[cls_id][:, 4:5] < self.conf_thres\n+ inds_low = pred_dets_dict[cls_id][:, 1:2] > self.low_conf_thres\n+ inds_high = pred_dets_dict[cls_id][:, 1:2] < self.conf_thres\ninds_second = np.logical_and(inds_low, inds_high).squeeze(-1)\npred_dets_cls_second = pred_dets_dict[cls_id][inds_second]\n# association the untrack to the low score detections\nif len(pred_dets_cls_second) > 0:\ndetections_second = [\n- STrack(STrack.tlbr_to_tlwh(tlbrs[:4]), tlbrs[4], cls_id, 30, temp_feat=None)\n+ STrack(\n+ STrack.tlbr_to_tlwh(tlbrs[:4]),\n+ tlbrs[4],\n+ cls_id,\n+ 30,\n+ temp_feat=None)\nfor tlbrs in pred_dets_cls_second[:, :5]\n]\nelse:\n@@ -222,7 +235,8 @@ class JDETracker(object):\ntrack_pool_dict[cls_id][i] for i in u_track\nif track_pool_dict[cls_id][i].state == TrackState.Tracked\n]\n- dists = matching.iou_distance(r_tracked_stracks, detections_second)\n+ dists = matching.iou_distance(r_tracked_stracks,\n+ detections_second)\nmatches, u_track, u_detection_second = matching.linear_assignment(\ndists, thresh=0.4) # not r_tracked_thresh\nelse:\n@@ -238,7 +252,8 @@ class JDETracker(object):\nfor i_tracked, idet in matches:\ntrack = r_tracked_stracks[i_tracked]\n- det = detections[idet] if not self.use_byte else detections_second[idet]\n+ det = detections[\n+ idet] if not self.use_byte else detections_second[idet]\nif track.state == TrackState.Tracked:\ntrack.update(det, self.frame_id)\nactivated_tracks_dict[cls_id].append(track)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/post_process.py",
"new_path": "ppdet/modeling/post_process.py",
"diff": "@@ -504,11 +504,10 @@ class CenterNetPostProcess(TTFBox):\nboxes_shape = bboxes.shape[:]\nscale_expand = paddle.expand(scale_expand, shape=boxes_shape)\nbboxes = paddle.divide(bboxes, scale_expand)\n+ results = paddle.concat([clses, scores, bboxes], axis=1)\nif self.for_mot:\n- results = paddle.concat([bboxes, scores, clses], axis=1)\nreturn results, inds, topk_clses\nelse:\n- results = paddle.concat([clses, scores, bboxes], axis=1)\nreturn results, paddle.shape(results)[0:1], topk_clses\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/reid/jde_embedding_head.py",
"new_path": "ppdet/modeling/reid/jde_embedding_head.py",
"diff": "@@ -152,9 +152,8 @@ class JDEEmbeddingHead(nn.Layer):\nscale_factor = targets['scale_factor'][0].numpy()\nbboxes[:, 2:] = self.scale_coords(bboxes[:, 2:], input_shape,\nim_shape, scale_factor)\n- # tlwhs, scores, cls_ids\n- pred_dets = paddle.concat(\n- (bboxes[:, 2:], bboxes[:, 1:2], bboxes[:, 0:1]), axis=1)\n+ # cls_ids, scores, tlwhs\n+ pred_dets = bboxes\nreturn pred_dets, pred_embs\ndef scale_coords(self, coords, input_shape, im_shape, scale_factor):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[MOT] unify mot and det output format (#5320)
|
499,339 |
09.03.2022 21:15:30
| -28,800 |
11c1efffc35ed2c6638abe62713b9b97700ec063
|
[SSD] add use_fuse_decode to SSDBox
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/layers.py",
"new_path": "ppdet/modeling/layers.py",
"diff": "@@ -553,9 +553,14 @@ class YOLOBox(object):\n@register\n@serializable\nclass SSDBox(object):\n- def __init__(self, is_normalized=True):\n+ def __init__(self,\n+ is_normalized=True,\n+ prior_box_var=[0.1, 0.1, 0.2, 0.2],\n+ use_fuse_decode=False):\nself.is_normalized = is_normalized\nself.norm_delta = float(not self.is_normalized)\n+ self.prior_box_var = prior_box_var\n+ self.use_fuse_decode = use_fuse_decode\ndef __call__(self,\npreds,\n@@ -564,40 +569,42 @@ class SSDBox(object):\nscale_factor,\nvar_weight=None):\nboxes, scores = preds\n- outputs = []\n- for box, score, prior_box in zip(boxes, scores, prior_boxes):\n- pb_w = prior_box[:, 2] - prior_box[:, 0] + self.norm_delta\n- pb_h = prior_box[:, 3] - prior_box[:, 1] + self.norm_delta\n- pb_x = prior_box[:, 0] + pb_w * 0.5\n- pb_y = prior_box[:, 1] + pb_h * 0.5\n- out_x = pb_x + box[:, :, 0] * pb_w * 0.1\n- out_y = pb_y + box[:, :, 1] * pb_h * 0.1\n- out_w = paddle.exp(box[:, :, 2] * 0.2) * pb_w\n- out_h = paddle.exp(box[:, :, 3] * 0.2) * pb_h\n-\n- if self.is_normalized:\n- h = paddle.unsqueeze(\n- im_shape[:, 0] / scale_factor[:, 0], axis=-1)\n- w = paddle.unsqueeze(\n- im_shape[:, 1] / scale_factor[:, 1], axis=-1)\n- output = paddle.stack(\n- [(out_x - out_w / 2.) * w, (out_y - out_h / 2.) * h,\n- (out_x + out_w / 2.) * w, (out_y + out_h / 2.) * h],\n- axis=-1)\n+ boxes = paddle.concat(boxes, axis=1)\n+ prior_boxes = paddle.concat(prior_boxes)\n+ if self.use_fuse_decode:\n+ output_boxes = ops.box_coder(\n+ prior_boxes,\n+ self.prior_box_var,\n+ boxes,\n+ code_type=\"decode_center_size\",\n+ box_normalized=self.is_normalized)\nelse:\n- output = paddle.stack(\n+ pb_w = prior_boxes[:, 2] - prior_boxes[:, 0] + self.norm_delta\n+ pb_h = prior_boxes[:, 3] - prior_boxes[:, 1] + self.norm_delta\n+ pb_x = prior_boxes[:, 0] + pb_w * 0.5\n+ pb_y = prior_boxes[:, 1] + pb_h * 0.5\n+ out_x = pb_x + boxes[:, :, 0] * pb_w * self.prior_box_var[0]\n+ out_y = pb_y + boxes[:, :, 1] * pb_h * self.prior_box_var[1]\n+ out_w = paddle.exp(boxes[:, :, 2] * self.prior_box_var[2]) * pb_w\n+ out_h = paddle.exp(boxes[:, :, 3] * self.prior_box_var[3]) * pb_h\n+ output_boxes = paddle.stack(\n[\n- out_x - out_w / 2., out_y - out_h / 2.,\n- out_x + out_w / 2. - 1., out_y + out_h / 2. - 1.\n+ out_x - out_w / 2., out_y - out_h / 2., out_x + out_w / 2.,\n+ out_y + out_h / 2.\n],\naxis=-1)\n- outputs.append(output)\n- boxes = paddle.concat(outputs, axis=1)\n- scores = F.softmax(paddle.concat(scores, axis=1))\n- scores = paddle.transpose(scores, [0, 2, 1])\n+ if self.is_normalized:\n+ h = (im_shape[:, 0] / scale_factor[:, 0]).unsqueeze(-1)\n+ w = (im_shape[:, 1] / scale_factor[:, 1]).unsqueeze(-1)\n+ im_shape = paddle.stack([w, h, w, h], axis=-1)\n+ output_boxes *= im_shape\n+ else:\n+ output_boxes[..., -2:] -= 1.0\n+ output_scores = F.softmax(paddle.concat(\n+ scores, axis=1)).transpose([0, 2, 1])\n- return boxes, scores\n+ return output_boxes, output_scores\n@register\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[SSD] add use_fuse_decode to SSDBox (#5336)
|
499,333 |
10.03.2022 15:16:15
| -28,800 |
80b1789eb3406f4f0d26290d4be48cecf766af73
|
add attr in deploy
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/python/attr_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 glob\n+from functools import reduce\n+\n+import cv2\n+import numpy as np\n+import math\n+import paddle\n+from paddle.inference import Config\n+from paddle.inference import create_predictor\n+\n+import sys\n+# add deploy path of PadleDetection to sys.path\n+parent_path = os.path.abspath(os.path.join(__file__, *(['..'])))\n+sys.path.insert(0, parent_path)\n+\n+from benchmark_utils import PaddleInferBenchmark\n+from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, WarpAffine\n+from visualize import visualize_attr\n+from utils import argsparser, Timer, get_current_memory_mb\n+from infer import Detector, get_test_images, print_arguments, load_predictor\n+\n+from PIL import Image, ImageDraw, ImageFont\n+\n+\n+class AttrDetector(Detector):\n+ \"\"\"\n+ Args:\n+ pred_config (object): config of model, defined by `Config(model_dir)`\n+ model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml\n+ device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU\n+ run_mode (str): mode of running(paddle/trt_fp32/trt_fp16)\n+ batch_size (int): size of pre batch in inference\n+ trt_min_shape (int): min shape for dynamic shape in trt\n+ trt_max_shape (int): max shape for dynamic shape in trt\n+ trt_opt_shape (int): opt shape for dynamic shape in trt\n+ trt_calib_mode (bool): If the model is produced by TRT offline quantitative\n+ calibration, trt_calib_mode need to set True\n+ cpu_threads (int): cpu threads\n+ enable_mkldnn (bool): whether to open MKLDNN\n+ output_dir (str): The path of output\n+ threshold (float): The threshold of score for visualization\n+ \"\"\"\n+\n+ def __init__(\n+ self,\n+ model_dir,\n+ device='CPU',\n+ run_mode='paddle',\n+ batch_size=1,\n+ trt_min_shape=1,\n+ trt_max_shape=1280,\n+ trt_opt_shape=640,\n+ trt_calib_mode=False,\n+ cpu_threads=1,\n+ enable_mkldnn=False,\n+ output_dir='output',\n+ threshold=0.5, ):\n+ super(AttrDetector, self).__init__(\n+ model_dir=model_dir,\n+ device=device,\n+ run_mode=run_mode,\n+ batch_size=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+ output_dir=output_dir,\n+ threshold=threshold, )\n+\n+ def get_label(self):\n+ return self.pred_config.labels\n+\n+ def postprocess(self, inputs, result):\n+ # postprocess output of predictor\n+ im_results = result['output']\n+ im_results = np.where(im_results < self.threshold, 0, im_results)\n+ label_list = [['Head', ['Hat', 'Glasses']], [\n+ 'Upper', [\n+ 'ShortSleeve', 'LongSleeve', 'UpperStride', 'UpperLogo',\n+ 'UpperPlaid', 'UpperSplice'\n+ ]\n+ ], [\n+ 'Lower', [\n+ 'LowerStripe', 'LowerPattern', 'LongCoat', 'Trousers', 'Shorts',\n+ 'Skirt&Dress'\n+ ]\n+ ], ['Shoes', ['boots']], [\n+ 'Accessory',\n+ ['HandBag', 'ShoulderBag', 'Backpack', 'HoldObjectsInFront']\n+ ], ['Age', ['AgeOver60', 'Age18-60', 'AgeLess18']],\n+ ['Gender', ['Female']],\n+ ['Direction', ['Front', 'Side', 'Back']]]\n+\n+ attr_type = [name[0] for name in label_list]\n+ labels = self.pred_config.labels\n+\n+ batch_res = []\n+ for res in im_results:\n+ label_res = {}\n+ label_res = {t: [] for t in attr_type}\n+ num = 0\n+ for i in range(len(label_list)):\n+ type_name_i = attr_type[i]\n+ attr_name_list = label_list[i][1]\n+ for attr_name in attr_name_list:\n+ attr_name = labels[num]\n+ output_prob = res[num]\n+ if output_prob != 0:\n+ label_res[type_name_i].append(attr_name)\n+ num += 1\n+\n+ if len(label_res['Shoes']) == 0:\n+ label_res['Shoes'] = ['no boots']\n+ if len(label_res['Gender']) == 0:\n+ label_res['Gender'] = ['Male']\n+ label_res['Age'] = [labels[19 + np.argmax(res[19:22])]]\n+ label_res['Direction'] = [labels[23 + np.argmax(res[23:])]]\n+ batch_res.append(label_res)\n+ result = {'output': batch_res}\n+ return result\n+\n+ def predict(self, repeats=1):\n+ '''\n+ Args:\n+ repeats (int): repeats number for prediction\n+ Returns:\n+ result (dict): include 'boxes': np.ndarray: shape:[N,6], N: number of box,\n+ matix element:[class, score, x_min, y_min, x_max, y_max]\n+ MaskRCNN's result include 'masks': np.ndarray:\n+ shape: [N, im_h, im_w]\n+ '''\n+ # model prediction\n+ for i in range(repeats):\n+ self.predictor.run()\n+ output_names = self.predictor.get_output_names()\n+ output_tensor = self.predictor.get_output_handle(output_names[0])\n+ np_output = output_tensor.copy_to_cpu()\n+ result = dict(output=np_output)\n+ return result\n+\n+ def predict_image(self,\n+ image_list,\n+ run_benchmark=False,\n+ repeats=1,\n+ visual=True):\n+ batch_loop_cnt = math.ceil(float(len(image_list)) / self.batch_size)\n+ results = []\n+ for i in range(batch_loop_cnt):\n+ start_index = i * self.batch_size\n+ end_index = min((i + 1) * self.batch_size, len(image_list))\n+ batch_image_list = image_list[start_index:end_index]\n+ if run_benchmark:\n+ # preprocess\n+ inputs = self.preprocess(batch_image_list) # warmup\n+ self.det_times.preprocess_time_s.start()\n+ inputs = self.preprocess(batch_image_list)\n+ self.det_times.preprocess_time_s.end()\n+\n+ # model prediction\n+ result = self.predict(repeats=repeats) # warmup\n+ self.det_times.inference_time_s.start()\n+ result = self.predict(repeats=repeats)\n+ self.det_times.inference_time_s.end(repeats=repeats)\n+\n+ # postprocess\n+ result_warmup = self.postprocess(inputs, result) # warmup\n+ self.det_times.postprocess_time_s.start()\n+ result = self.postprocess(inputs, result)\n+ self.det_times.postprocess_time_s.end()\n+ self.det_times.img_num += len(batch_image_list)\n+\n+ cm, gm, gu = get_current_memory_mb()\n+ self.cpu_mem += cm\n+ self.gpu_mem += gm\n+ self.gpu_util += gu\n+ else:\n+ # preprocess\n+ self.det_times.preprocess_time_s.start()\n+ inputs = self.preprocess(batch_image_list)\n+ self.det_times.preprocess_time_s.end()\n+\n+ # model prediction\n+ self.det_times.inference_time_s.start()\n+ result = self.predict()\n+ self.det_times.inference_time_s.end()\n+\n+ # postprocess\n+ self.det_times.postprocess_time_s.start()\n+ result = self.postprocess(inputs, result)\n+ self.det_times.postprocess_time_s.end()\n+ self.det_times.img_num += len(batch_image_list)\n+\n+ if visual:\n+ visualize(\n+ batch_image_list, result, output_dir=self.output_dir)\n+\n+ results.append(result)\n+ if visual:\n+ print('Test iter {}'.format(i))\n+\n+ results = self.merge_batch_result(results)\n+ return results\n+\n+ def merge_batch_result(self, batch_result):\n+ if len(batch_result) == 1:\n+ return batch_result[0]\n+ res_key = batch_result[0].keys()\n+ results = {k: [] for k in res_key}\n+ for res in batch_result:\n+ for k, v in res.items():\n+ results[k].extend(v)\n+ return results\n+\n+\n+def visualize(image_list, batch_res, output_dir='output'):\n+\n+ # visualize the predict result\n+ batch_res = batch_res['output']\n+ for image_file, res in zip(image_list, batch_res):\n+ im = visualize_attr(image_file, [res])\n+ if not os.path.exists(output_dir):\n+ os.makedirs(output_dir)\n+ img_name = os.path.split(image_file)[-1]\n+ out_path = os.path.join(output_dir, img_name)\n+ im.save(out_path, quality=95)\n+ print(\"save result to: \" + out_path)\n+\n+\n+def main():\n+ detector = AttrDetector(\n+ FLAGS.model_dir,\n+ device=FLAGS.device,\n+ run_mode=FLAGS.run_mode,\n+ batch_size=FLAGS.batch_size,\n+ trt_min_shape=FLAGS.trt_min_shape,\n+ trt_max_shape=FLAGS.trt_max_shape,\n+ trt_opt_shape=FLAGS.trt_opt_shape,\n+ trt_calib_mode=FLAGS.trt_calib_mode,\n+ cpu_threads=FLAGS.cpu_threads,\n+ enable_mkldnn=FLAGS.enable_mkldnn,\n+ threshold=FLAGS.threshold,\n+ output_dir=FLAGS.output_dir)\n+\n+ # predict from image\n+ if FLAGS.image_dir is None and FLAGS.image_file is not None:\n+ assert FLAGS.batch_size == 1, \"batch_size should be 1, when image_file is not None\"\n+ img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)\n+ detector.predict_image(img_list, FLAGS.run_benchmark, repeats=10)\n+ if not FLAGS.run_benchmark:\n+ detector.det_times.info(average=True)\n+ else:\n+ mems = {\n+ 'cpu_rss_mb': detector.cpu_mem / len(img_list),\n+ 'gpu_rss_mb': detector.gpu_mem / len(img_list),\n+ 'gpu_util': detector.gpu_util * 100 / len(img_list)\n+ }\n+\n+ perf_info = detector.det_times.report(average=True)\n+ model_dir = FLAGS.model_dir\n+ mode = FLAGS.run_mode\n+ model_info = {\n+ 'model_name': model_dir.strip('/').split('/')[-1],\n+ 'precision': mode.split('_')[-1]\n+ }\n+ data_info = {\n+ 'batch_size': FLAGS.batch_size,\n+ 'shape': \"dynamic_shape\",\n+ 'data_num': perf_info['img_num']\n+ }\n+ det_log = PaddleInferBenchmark(detector.config, model_info, data_info,\n+ perf_info, mems)\n+ det_log('Attr')\n+\n+\n+if __name__ == '__main__':\n+ paddle.enable_static()\n+ parser = argsparser()\n+ FLAGS = parser.parse_args()\n+ print_arguments(FLAGS)\n+ FLAGS.device = FLAGS.device.upper()\n+ assert FLAGS.device in ['CPU', 'GPU', 'XPU'\n+ ], \"device should be CPU, GPU or XPU\"\n+ assert not FLAGS.use_gpu, \"use_gpu has been deprecated, please use --device\"\n+\n+ main()\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add attr in deploy (#5342)
|
499,304 |
11.03.2022 18:11:44
| -28,800 |
f08c2ca7aff0eec9bb590d861c37a6700b7ce2ea
|
fix picoheadv2 use_align_head
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/pico_head.py",
"new_path": "ppdet/modeling/heads/pico_head.py",
"diff": "@@ -466,9 +466,7 @@ class PicoHeadV2(GFLHead):\n), \"The size of fpn_feats is not equal to size of fpn_stride\"\ncls_score_list, reg_list, box_list = [], [], []\n- for i, fpn_feat, stride, align_cls in zip(\n- range(len(self.fpn_stride)), fpn_feats, self.fpn_stride,\n- self.cls_align):\n+ for i, (fpn_feat, stride) in enumerate(zip(fpn_feats, self.fpn_stride)):\nb, _, h, w = get_static_shape(fpn_feat)\n# task decomposition\nconv_cls_feat, se_feat = self.conv_feat(fpn_feat, i)\n@@ -477,7 +475,7 @@ class PicoHeadV2(GFLHead):\n# cls prediction and alignment\nif self.use_align_head:\n- cls_prob = F.sigmoid(align_cls(conv_cls_feat))\n+ cls_prob = F.sigmoid(self.cls_align[i](conv_cls_feat))\ncls_score = (F.sigmoid(cls_logit) * cls_prob + eps).sqrt()\nelse:\ncls_score = F.sigmoid(cls_logit)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix picoheadv2 use_align_head (#5352)
|
499,373 |
15.03.2022 20:34:38
| -28,800 |
1acb18522ac110fe38f308600c176bb814ee6d2f
|
[Fix] Update operators.py
Fixed a possible error on line 2021: "TypeError: 'numpy.float64' object cannot be interpreted as an integer".
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/transform/operators.py",
"new_path": "ppdet/data/transform/operators.py",
"diff": "@@ -2032,8 +2032,8 @@ class Pad(BaseOperator):\nim_h < h and im_w < w\n), '(h, w) of target size should be greater than (im_h, im_w)'\nelse:\n- h = np.ceil(im_h / self.size_divisor) * self.size_divisor\n- w = np.ceil(im_w / self.size_divisor) * self.size_divisor\n+ h = int(np.ceil(im_h / self.size_divisor) * self.size_divisor)\n+ w = int(np.ceil(im_w / self.size_divisor) * self.size_divisor)\nif h == im_h and w == im_w:\nreturn sample\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[Fix] Update operators.py (#5357)
Fixed a possible error on line 2021: "TypeError: 'numpy.float64' object cannot be interpreted as an integer".
|
499,339 |
15.03.2022 21:00:45
| -28,800 |
3d1dfea9eb39b09b149ee8d097aec732a76f2d25
|
[dev] add use_epoch in LinearWarmup
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer.py",
"diff": "@@ -132,19 +132,24 @@ class LinearWarmup(object):\nArgs:\nsteps (int): warm up steps\nstart_factor (float): initial learning rate factor\n+ epochs (int|None): use epochs as warm up steps, the priority\n+ of `epochs` is higher than `steps`. Default: None.\n\"\"\"\n- def __init__(self, steps=500, start_factor=1. / 3):\n+ def __init__(self, steps=500, start_factor=1. / 3, epochs=None):\nsuper(LinearWarmup, self).__init__()\nself.steps = steps\nself.start_factor = start_factor\n+ self.epochs = epochs\ndef __call__(self, base_lr, step_per_epoch):\nboundary = []\nvalue = []\n- for i in range(self.steps + 1):\n- if self.steps > 0:\n- alpha = i / self.steps\n+ warmup_steps = self.epochs * step_per_epoch \\\n+ if self.epochs is not None else self.steps\n+ for i in range(warmup_steps + 1):\n+ if warmup_steps > 0:\n+ alpha = i / warmup_steps\nfactor = self.start_factor * (1 - alpha) + alpha\nlr = base_lr * factor\nvalue.append(lr)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[dev] add use_epoch in LinearWarmup (#5366)
|
499,304 |
18.03.2022 11:16:37
| -28,800 |
50da62fdf5530437e83dfb39bca8f278f64cecf2
|
fix picodet_pedestrian export
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/application/pedestrian_detection/picodet_s_192_pedestrian.yml",
"new_path": "configs/picodet/application/pedestrian_detection/picodet_s_192_pedestrian.yml",
"diff": "@@ -12,6 +12,11 @@ snapshot_epoch: 10\nepoch: 300\nmetric: COCO\nnum_classes: 1\n+# Exporting the model\n+export:\n+ post_process: False # Whether post-processing is included in the network when export model.\n+ nms: False # Whether NMS is included in the network when export model.\n+ benchmark: False # It is used to testing model performance, if set `True`, post-process and NMS will not be exported.\narchitecture: PicoDet\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/picodet/application/pedestrian_detection/picodet_s_320_pedestrian.yml",
"new_path": "configs/picodet/application/pedestrian_detection/picodet_s_320_pedestrian.yml",
"diff": "@@ -12,6 +12,11 @@ snapshot_epoch: 10\nepoch: 300\nmetric: COCO\nnum_classes: 1\n+# Exporting the model\n+export:\n+ post_process: False # Whether post-processing is included in the network when export model.\n+ nms: False # Whether NMS is included in the network when export model.\n+ benchmark: False # It is used to testing model performance, if set `True`, post-process and NMS will not be exported.\narchitecture: PicoDet\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix picodet_pedestrian export (#5395)
|
499,395 |
19.03.2022 20:38:52
| -28,800 |
911112265a99b17f487f4902e1e979f0dc784d7f
|
fix register_buffer
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/bbox_utils.py",
"new_path": "ppdet/modeling/bbox_utils.py",
"diff": "@@ -763,8 +763,9 @@ def batch_distance2bbox(points, distance, max_shapes=None):\nTensor: Decoded bboxes, \"x1y1x2y2\" format.\n\"\"\"\nlt, rb = paddle.split(distance, 2, -1)\n- x1y1 = points - lt\n- x2y2 = points + rb\n+ # while tensor add parameters, parameters should be better placed on the second place\n+ x1y1 = -lt + points\n+ x2y2 = rb + points\nout_bbox = paddle.concat([x1y1, x2y2], -1)\nif max_shapes is not None:\nmax_shapes = max_shapes.flip(-1).tile([1, 2])\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix register_buffer (#5398)
|
499,312 |
21.03.2022 14:24:02
| -28,800 |
6a17524f8377eed56ac1713877cf43ea18a02f94
|
adjust blazeface's search space
|
[
{
"change_type": "MODIFY",
"old_path": "static/slim/nas/search_space/blazefacespace_nas.py",
"new_path": "static/slim/nas/search_space/blazefacespace_nas.py",
"diff": "@@ -28,8 +28,8 @@ class BlazeFaceNasSpace(SearchSpaceBase):\ndef __init__(self, input_size, output_size, block_num, block_mask):\nsuper(BlazeFaceNasSpace, self).__init__(input_size, output_size,\nblock_num, block_mask)\n- self.blaze_filter_num1 = np.array([4, 8, 12, 16, 20, 24, 32])\n- self.blaze_filter_num2 = np.array([8, 12, 16, 20, 24, 32, 40, 48, 64])\n+ self.blaze_filter_num1 = np.array([4, 8, 12, 16, 24, 32])\n+ self.blaze_filter_num2 = np.array([8, 12, 16, 24, 32, 40, 48, 64])\nself.mid_filter_num = np.array([8, 12, 16, 20, 24, 32])\nself.double_filter_num = np.array(\n[8, 12, 16, 24, 32, 40, 48, 64, 72, 80, 88, 96])\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
adjust blazeface's search space (#5404)
|
499,304 |
21.03.2022 19:27:54
| -28,800 |
c616c10c7b7745915892bb055c916c7179b2b50a
|
fix new PicoDet name
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/legacy_model/README.md",
"new_path": "configs/picodet/legacy_model/README.md",
"diff": "- PicoDet used 4 or 8 GPUs for training and all checkpoints are trained with default settings and hyperparameters.\n</details>\n+\n+## Cite PP-PicoDet\n+```\n+@misc{yu2021pppicodet,\n+ title={PP-PicoDet: A Better Real-Time Object Detector on Mobile Devices},\n+ author={Guanghua Yu and Qinyao Chang and Wenyu Lv and Chang Xu and Cheng Cui and Wei Ji and Qingqing Dang and Kaipeng Deng and Guanzhong Wang and Yuning Du and Baohua Lai and Qiwen Liu and Xiaoguang Hu and Dianhai Yu and Yanjun Ma},\n+ year={2021},\n+ eprint={2111.00902},\n+ archivePrefix={arXiv},\n+ primaryClass={cs.CV}\n+}\n+\n+```\n"
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_l_320_coco.yml",
"new_path": "configs/picodet/picodet_l_320_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_l_416_coco.yml",
"new_path": "configs/picodet/picodet_l_416_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_l_640_coco.yml",
"new_path": "configs/picodet/picodet_l_640_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_m_320_coco.yml",
"new_path": "configs/picodet/picodet_m_320_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_m_416_coco.yml",
"new_path": "configs/picodet/picodet_m_416_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_s_320_coco.yml",
"new_path": "configs/picodet/picodet_s_320_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_s_416_coco.yml",
"new_path": "configs/picodet/picodet_s_416_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_xs_320_coco.yml",
"new_path": "configs/picodet/picodet_xs_320_coco_lcnet.yml",
"diff": ""
},
{
"change_type": "RENAME",
"old_path": "configs/picodet/picodet_xs_416_coco.yml",
"new_path": "configs/picodet/picodet_xs_416_coco_lcnet.yml",
"diff": ""
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix new PicoDet name (#5414)
|
499,333 |
21.03.2022 22:13:31
| -28,800 |
8622966f6d09e1969ecf96534f8bf4bc1c2efd8f
|
update pphuman for pptracking
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/config/tracker_config.yml",
"new_path": "deploy/pphuman/config/tracker_config.yml",
"diff": "-# config of tracker for MOT SDE Detector, use ByteTracker as default.\n-# The tracker of MOT JDE Detector is exported together with the model.\n+# config of tracker for MOT SDE Detector, use 'JDETracker' as default.\n+# The tracker of MOT JDE Detector (such as FairMOT) is exported together with the model.\n# Here 'min_box_area' and 'vertical_ratio' are set for pedestrian, you can modify for other objects tracking.\n-tracker:\n- use_byte: true\n+\n+type: JDETracker # 'JDETracker' or 'DeepSORTTracker'\n+\n+# BYTETracker\n+JDETracker:\n+ use_byte: True\n+ det_thresh: 0.3\nconf_thres: 0.6\nlow_conf_thres: 0.1\nmatch_thres: 0.9\nmin_box_area: 100\n- vertical_ratio: 1.6\n+ vertical_ratio: 1.6 # for pedestrian\n+\n+DeepSORTTracker:\n+ input_size: [64, 192]\n+ min_box_area: 0\n+ vertical_ratio: -1\n+ budget: 100\n+ max_age: 70\n+ n_init: 3\n+ metric_type: cosine\n+ matching_threshold: 0.2\n+ max_iou_distance: 0.9\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -28,7 +28,6 @@ parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))\nsys.path.insert(0, parent_path)\nfrom python.infer import Detector, DetectorPicoDet\n-from python.mot_sde_infer import SDE_Detector\nfrom python.attr_infer import AttrDetector\nfrom python.keypoint_infer import KeyPointDetector\nfrom python.keypoint_postprocess import translate_to_ori_images\n@@ -39,6 +38,8 @@ from 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\nfrom python.preprocess import decode_image\nfrom python.visualize import visualize_box_mask, visualize_attr, visualize_pose, visualize_action\n+\n+from pptracking.python.mot_sde_infer import SDE_Detector\nfrom pptracking.python.visualize import plot_tracking\n@@ -374,6 +375,8 @@ class PipePredictor(object):\n# det output format: class, score, xmin, ymin, xmax, ymax\ndet_res = self.det_predictor.predict_image(\nbatch_input, visual=False)\n+ det_res = self.det_predictor.filter_box(det_res,\n+ self.cfg['crop_thresh'])\nif i > self.warmup_frame:\nself.pipe_timer.module_time['det'].end()\nself.pipeline_res.update(det_res, 'det')\n@@ -563,6 +566,8 @@ class PipePredictor(object):\ndet_res_i,\nlabels=['person'],\nthreshold=self.cfg['crop_thresh'])\n+ im = np.ascontiguousarray(np.copy(im))\n+ im = cv2.cvtColor(im, cv2.COLOR_RGB2BGR)\nif attr_res is not None:\nattr_res_i = attr_res['output'][start_idx:start_idx +\nboxes_num_i]\n@@ -571,7 +576,7 @@ class PipePredictor(object):\nif not os.path.exists(self.output_dir):\nos.makedirs(self.output_dir)\nout_path = os.path.join(self.output_dir, img_name)\n- im.save(out_path, quality=95)\n+ cv2.imwrite(out_path, im)\nprint(\"save result to: \" + out_path)\nstart_idx += boxes_num_i\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot_sde_infer.py",
"new_path": "deploy/pptracking/python/mot_sde_infer.py",
"diff": "@@ -24,8 +24,8 @@ import paddle\nfrom benchmark_utils import PaddleInferBenchmark\nfrom preprocess import decode_image\n-from utils import argsparser, Timer, get_current_memory_mb, _is_valid_video, video2frames\n-from det_infer import Detector, get_test_images, print_arguments, bench_log, PredictConfig, load_predictor\n+from .utils import argsparser, Timer, get_current_memory_mb, _is_valid_video, video2frames\n+from .det_infer import Detector, get_test_images, print_arguments, bench_log, PredictConfig, load_predictor\n# add python path\nimport sys\n@@ -34,7 +34,7 @@ sys.path.insert(0, parent_path)\nfrom mot.tracker import JDETracker, DeepSORTTracker\nfrom mot.utils import MOTTimer, write_mot_results, flow_statistic, get_crops, clip_box\n-from visualize import plot_tracking, plot_tracking_dict\n+from .visualize import plot_tracking, plot_tracking_dict\nfrom mot.mtmct.utils import parse_bias\nfrom mot.mtmct.postprocess import trajectory_fusion, sub_cluster, gen_res, print_mtmct_result\n@@ -120,10 +120,12 @@ class SDE_Detector(Detector):\ncfg = tracker_cfg[tracker_cfg['type']]\n# tracker config\n- self.use_deepsort_tracker = True if tracker_cfg['type'] == 'DeepSORTTracker' else False\n+ self.use_deepsort_tracker = True if tracker_cfg[\n+ 'type'] == 'DeepSORTTracker' else False\nif self.use_deepsort_tracker:\n# use DeepSORTTracker\n- if self.reid_pred_config is not None and hasattr(self.reid_pred_config, 'tracker'):\n+ if self.reid_pred_config is not None and hasattr(\n+ self.reid_pred_config, 'tracker'):\ncfg = self.reid_pred_config.tracker\nbudget = cfg.get('budget', 100)\nmax_age = cfg.get('max_age', 30)\n@@ -138,8 +140,7 @@ class SDE_Detector(Detector):\nmax_iou_distance=max_iou_distance,\nmatching_threshold=matching_threshold,\nmin_box_area=min_box_area,\n- vertical_ratio=vertical_ratio,\n- )\n+ vertical_ratio=vertical_ratio, )\nelse:\n# use ByteTracker\nuse_byte = cfg.get('use_byte', False)\n@@ -158,8 +159,7 @@ class SDE_Detector(Detector):\nvertical_ratio=vertical_ratio,\nmatch_thres=match_thres,\nconf_thres=conf_thres,\n- low_conf_thres=low_conf_thres,\n- )\n+ low_conf_thres=low_conf_thres, )\nself.do_mtmct = False if mtmct_dir is None else True\nself.mtmct_dir = mtmct_dir\n@@ -206,7 +206,8 @@ class SDE_Detector(Detector):\nfor i in range(repeats):\nself.reid_predictor.run()\noutput_names = self.reid_predictor.get_output_names()\n- feature_tensor = self.reid_predictor.get_output_handle(output_names[0])\n+ feature_tensor = self.reid_predictor.get_output_handle(output_names[\n+ 0])\npred_embs = feature_tensor.copy_to_cpu()\ndet_results['embeddings'] = pred_embs\n@@ -249,7 +250,8 @@ class SDE_Detector(Detector):\nframe_id = det_results['frame_id']\ntracking_outs['feat_data'] = {}\n- for _tlbr, _id, _feat in zip(online_tlbrs, online_ids, online_feats):\n+ for _tlbr, _id, _feat in zip(online_tlbrs, online_ids,\n+ online_feats):\nfeat_data = {}\nfeat_data['bbox'] = _tlbr\nfeat_data['frame'] = f\"{frame_id:06d}\"\n@@ -265,7 +267,8 @@ class SDE_Detector(Detector):\nonline_scores = defaultdict(list)\nonline_ids = defaultdict(list)\nif self.do_mtmct:\n- online_tlbrs, online_feats = defaultdict(list), defaultdict(list)\n+ online_tlbrs, online_feats = defaultdict(list), defaultdict(\n+ list)\nonline_targets_dict = self.tracker.update(pred_dets, pred_embs)\nfor cls_id in range(self.num_classes):\nonline_targets = online_targets_dict[cls_id]\n@@ -295,7 +298,8 @@ class SDE_Detector(Detector):\nseq_name = det_results['seq_name']\nframe_id = det_results['frame_id']\ntracking_outs['feat_data'] = {}\n- for _tlbr, _id, _feat in zip(online_tlbrs[0], online_ids[0], online_feats[0]):\n+ for _tlbr, _id, _feat in zip(online_tlbrs[0], online_ids[0],\n+ online_feats[0]):\nfeat_data = {}\nfeat_data['bbox'] = _tlbr\nfeat_data['frame'] = f\"{frame_id:06d}\"\n@@ -469,14 +473,17 @@ class SDE_Detector(Detector):\ntimer.tic()\nseq_name = video_out_name.split('.')[0]\n- mot_results = self.predict_image([frame], visual=False, seq_name=seq_name)\n+ mot_results = self.predict_image(\n+ [frame], visual=False, seq_name=seq_name)\ntimer.toc()\n- online_tlwhs, online_scores, online_ids = mot_results[0] # bs=1 in MOT model\n+ online_tlwhs, online_scores, online_ids = mot_results[\n+ 0] # bs=1 in MOT model\nfps = 1. / timer.duration\nif num_classes == 1 and self.use_reid:\n# use DeepSORTTracker, only support singe class\n- results[0].append((frame_id + 1, online_tlwhs, online_scores, online_ids))\n+ results[0].append(\n+ (frame_id + 1, online_tlwhs, online_scores, online_ids))\nim = plot_tracking(\nframe,\nonline_tlwhs,\n@@ -488,8 +495,8 @@ class SDE_Detector(Detector):\n# use ByteTracker, support multiple class\nfor cls_id in range(num_classes):\nresults[cls_id].append(\n- (frame_id + 1, online_tlwhs[cls_id], online_scores[cls_id],\n- online_ids[cls_id]))\n+ (frame_id + 1, online_tlwhs[cls_id],\n+ online_scores[cls_id], online_ids[cls_id]))\nim = plot_tracking_dict(\nframe,\nnum_classes,\n@@ -549,13 +556,15 @@ class SDE_Detector(Detector):\ncontinue\nif os.path.exists(os.path.join(fpath, 'img1')):\nfpath = os.path.join(fpath, 'img1')\n- assert os.path.isdir(fpath), '{} should be a directory'.format(fpath)\n+ assert os.path.isdir(fpath), '{} should be a directory'.format(\n+ fpath)\nimage_list = glob.glob(os.path.join(fpath, '*.jpg'))\nimage_list.sort()\nassert len(image_list) > 0, '{} has no images.'.format(fpath)\nprint('start tracking seq: {}'.format(seq))\n- mot_features_dict = self.predict_image(image_list, visual=False, seq_name=seq)\n+ mot_features_dict = self.predict_image(\n+ image_list, visual=False, seq_name=seq)\ncid = int(re.sub('[a-z,A-Z]', \"\", seq))\ntid_data, mot_list_break = trajectory_fusion(\n@@ -627,8 +636,7 @@ def main():\nthreshold=FLAGS.threshold,\noutput_dir=FLAGS.output_dir,\nreid_model_dir=FLAGS.reid_model_dir,\n- mtmct_dir=FLAGS.mtmct_dir,\n- )\n+ mtmct_dir=FLAGS.mtmct_dir, )\n# predict from video file or camera video stream\nif FLAGS.video_file is not None or FLAGS.camera_id != -1:\n@@ -643,7 +651,8 @@ def main():\nassert FLAGS.batch_size == 1, \"--batch_size should be 1 in MOT models.\"\nimg_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)\nseq_name = FLAGS.image_dir.split('/')[-1]\n- detector.predict_image(img_list, FLAGS.run_benchmark, repeats=10, seq_name=seq_name)\n+ detector.predict_image(\n+ img_list, FLAGS.run_benchmark, repeats=10, seq_name=seq_name)\nif not FLAGS.run_benchmark:\ndetector.det_times.info(average=True)\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -150,6 +150,25 @@ class Detector(object):\nresult = {k: v for k, v in result.items() if v is not None}\nreturn result\n+ def filter_box(self, result, threshold):\n+ np_boxes_num = result['boxes_num']\n+ boxes = result['boxes']\n+ start_idx = 0\n+ filter_boxes = []\n+ filter_num = []\n+ for i in range(len(np_boxes_num)):\n+ boxes_num = np_boxes_num[i]\n+ boxes_i = boxes[start_idx:start_idx + boxes_num, :]\n+ idx = boxes_i[:, 1] > threshold\n+ filter_boxes_i = boxes_i[idx, :]\n+ filter_boxes.append(filter_boxes_i)\n+ filter_num.append(filter_boxes_i.shape[0])\n+ start_idx += boxes_num\n+ boxes = np.concatenate(filter_boxes)\n+ filter_num = np.array(filter_num)\n+ filter_res = {'boxes': boxes, 'boxes_num': filter_num}\n+ return filter_res\n+\ndef predict(self, repeats=1):\n'''\nArgs:\n@@ -736,7 +755,8 @@ def main():\nelif arch == 'PicoDet':\ndetector_func = 'DetectorPicoDet'\n- detector = eval(detector_func)(FLAGS.model_dir,\n+ detector = eval(detector_func)(\n+ FLAGS.model_dir,\ndevice=FLAGS.device,\nrun_mode=FLAGS.run_mode,\nbatch_size=FLAGS.batch_size,\n@@ -781,6 +801,8 @@ if __name__ == '__main__':\n], \"device should be CPU, GPU or XPU\"\nassert not FLAGS.use_gpu, \"use_gpu has been deprecated, please use --device\"\n- assert not (FLAGS.enable_mkldnn==False and FLAGS.enable_mkldnn_bfloat16==True), 'To enable mkldnn bfloat, please turn on both enable_mkldnn and enable_mkldnn_bfloat16'\n+ assert not (\n+ FLAGS.enable_mkldnn == False and FLAGS.enable_mkldnn_bfloat16 == True\n+ ), 'To enable mkldnn bfloat, please turn on both enable_mkldnn and enable_mkldnn_bfloat16'\nmain()\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update pphuman for pptracking (#5419)
|
499,333 |
22.03.2022 12:23:43
| -28,800 |
227368762aa666c18dbb5ba06346490439e55188
|
refine pptracking import
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -40,7 +40,7 @@ from python.preprocess import decode_image\nfrom python.visualize import visualize_box_mask, visualize_attr, visualize_pose, visualize_action\nfrom pptracking.python.mot_sde_infer import SDE_Detector\n-from pptracking.python.visualize import plot_tracking\n+from pptracking.python.mot.visualize import plot_tracking\nclass Pipeline(object):\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/det_infer.py",
"new_path": "deploy/pptracking/python/det_infer.py",
"diff": "@@ -33,8 +33,8 @@ sys.path.insert(0, parent_path)\nfrom benchmark_utils import PaddleInferBenchmark\nfrom picodet_postprocess import PicoDetPostProcess\nfrom preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, decode_image\n-from visualize import visualize_box_mask\n-from utils import argsparser, Timer, get_current_memory_mb\n+from mot.visualize import visualize_box_mask\n+from mot_utils import argsparser, Timer, get_current_memory_mb\n# Global dictionary\nSUPPORT_MODELS = {\n"
},
{
"change_type": "RENAME",
"old_path": "deploy/pptracking/python/visualize.py",
"new_path": "deploy/pptracking/python/mot/visualize.py",
"diff": ""
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot_jde_infer.py",
"new_path": "deploy/pptracking/python/mot_jde_infer.py",
"diff": "@@ -22,7 +22,7 @@ import paddle\nfrom benchmark_utils import PaddleInferBenchmark\nfrom preprocess import decode_image\n-from utils import argsparser, Timer, get_current_memory_mb\n+from mot_utils import argsparser, Timer, get_current_memory_mb\nfrom det_infer import Detector, get_test_images, print_arguments, bench_log, PredictConfig\n# add python path\n@@ -31,8 +31,8 @@ parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))\nsys.path.insert(0, parent_path)\nfrom mot import JDETracker\n-from utils import MOTTimer, write_mot_results\n-from visualize import plot_tracking, plot_tracking_dict\n+from mot.utils import MOTTimer, write_mot_results\n+from mot.visualize import plot_tracking, plot_tracking_dict\n# Global dictionary\nMOT_JDE_SUPPORT_MODELS = {\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot_sde_infer.py",
"new_path": "deploy/pptracking/python/mot_sde_infer.py",
"diff": "@@ -24,17 +24,17 @@ import paddle\nfrom benchmark_utils import PaddleInferBenchmark\nfrom preprocess import decode_image\n-from .utils import argsparser, Timer, get_current_memory_mb, _is_valid_video, video2frames\n-from .det_infer import Detector, get_test_images, print_arguments, bench_log, PredictConfig, load_predictor\n# add python path\nimport sys\n-parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))\n+parent_path = os.path.abspath(os.path.join(__file__, *(['..'])))\nsys.path.insert(0, parent_path)\n+from det_infer import Detector, get_test_images, print_arguments, bench_log, PredictConfig, load_predictor\n+from mot_utils import argsparser, Timer, get_current_memory_mb, video2frames, _is_valid_video\nfrom mot.tracker import JDETracker, DeepSORTTracker\nfrom mot.utils import MOTTimer, write_mot_results, flow_statistic, get_crops, clip_box\n-from .visualize import plot_tracking, plot_tracking_dict\n+from mot.visualize import plot_tracking, plot_tracking_dict\nfrom mot.mtmct.utils import parse_bias\nfrom mot.mtmct.postprocess import trajectory_fusion, sub_cluster, gen_res, print_mtmct_result\n"
},
{
"change_type": "RENAME",
"old_path": "deploy/pptracking/python/utils.py",
"new_path": "deploy/pptracking/python/mot_utils.py",
"diff": "@@ -147,7 +147,6 @@ def argsparser():\nreturn parser\n-\nclass Times(object):\ndef __init__(self):\nself.time = 0.\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/mot_sde_infer.py",
"new_path": "deploy/python/mot_sde_infer.py",
"diff": "@@ -32,7 +32,7 @@ sys.path.insert(0, parent_path)\nfrom pptracking.python.mot import JDETracker\nfrom pptracking.python.mot.utils import MOTTimer, write_mot_results\n-from pptracking.python.visualize import plot_tracking, plot_tracking_dict\n+from pptracking.python.mot.visualize import plot_tracking, plot_tracking_dict\nclass SDE_Detector(Detector):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
refine pptracking import (#5421)
|
499,333 |
22.03.2022 20:28:16
| -28,800 |
e0a8d481843c5b900067664de3a6f0722f45f465
|
refine pphuman vis
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -521,13 +521,16 @@ class PipePredictor(object):\nmot_res = result.get('mot')\nif mot_res is not None:\nids = mot_res['boxes'][:, 0]\n+ scores = mot_res['boxes'][:, 2]\nboxes = mot_res['boxes'][:, 3:]\nboxes[:, 2] = boxes[:, 2] - boxes[:, 0]\nboxes[:, 3] = boxes[:, 3] - boxes[:, 1]\nelse:\nboxes = np.zeros([0, 4])\nids = np.zeros([0])\n- image = plot_tracking(image, boxes, ids, frame_id=frame_id, fps=fps)\n+ scores = np.zeros([0])\n+ image = plot_tracking(\n+ image, boxes, ids, scores, frame_id=frame_id, fps=fps)\nattr_res = result.get('attr')\nif attr_res is not None:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot/visualize.py",
"new_path": "deploy/pptracking/python/mot/visualize.py",
"diff": "@@ -134,47 +134,45 @@ def plot_tracking(image,\nim = np.ascontiguousarray(np.copy(image))\nim_h, im_w = im.shape[:2]\n- text_scale = max(1, image.shape[1] / 1600.)\n+ text_scale = max(0.5, image.shape[1] / 3000.)\ntext_thickness = 2\nline_thickness = max(1, int(image.shape[1] / 500.))\ncv2.putText(\nim,\n'frame: %d fps: %.2f num: %d' % (frame_id, fps, len(tlwhs)),\n- (0, int(15 * text_scale)),\n- cv2.FONT_HERSHEY_PLAIN,\n+ (0, int(15 * text_scale) + 5),\n+ cv2.FONT_ITALIC,\ntext_scale, (0, 0, 255),\n- thickness=2)\n-\n+ thickness=text_thickness)\nfor i, tlwh in enumerate(tlwhs):\nx1, y1, w, h = tlwh\nintbox = tuple(map(int, (x1, y1, x1 + w, y1 + h)))\nobj_id = int(obj_ids[i])\n- id_text = '{}'.format(int(obj_id))\n+ id_text = 'ID: {}'.format(int(obj_id))\nif ids2names != []:\nassert len(\nids2names) == 1, \"plot_tracking only supports single classes.\"\n- id_text = '{}_'.format(ids2names[0]) + id_text\n+ id_text = 'ID: {}_'.format(ids2names[0]) + id_text\n_line_thickness = 1 if obj_id <= 0 else line_thickness\ncolor = get_color(abs(obj_id))\ncv2.rectangle(\nim, intbox[0:2], intbox[2:4], color=color, thickness=line_thickness)\ncv2.putText(\nim,\n- id_text, (intbox[0], intbox[1] - 10),\n- cv2.FONT_HERSHEY_PLAIN,\n- text_scale, (0, 0, 255),\n+ id_text, (intbox[0], intbox[1] - 25),\n+ cv2.FONT_ITALIC,\n+ text_scale, (0, 255, 255),\nthickness=text_thickness)\nif scores is not None:\n- text = '{:.2f}'.format(float(scores[i]))\n+ text = 'score: {:.2f}'.format(float(scores[i]))\ncv2.putText(\nim,\n- text, (intbox[0], intbox[1] + 10),\n- cv2.FONT_HERSHEY_PLAIN,\n- text_scale, (0, 255, 255),\n+ text, (intbox[0], intbox[1] - 6),\n+ cv2.FONT_ITALIC,\n+ text_scale, (0, 255, 0),\nthickness=text_thickness)\n-\nif do_entrance_counting:\nentrance_line = tuple(map(int, entrance))\ncv2.rectangle(\n@@ -201,7 +199,7 @@ def plot_tracking_dict(image,\nim = np.ascontiguousarray(np.copy(image))\nim_h, im_w = im.shape[:2]\n- text_scale = max(1, image.shape[1] / 1600.)\n+ text_scale = max(0.5, image.shape[1] / 3000.)\ntext_thickness = 2\nline_thickness = max(1, int(image.shape[1] / 500.))\n@@ -212,9 +210,9 @@ def plot_tracking_dict(image,\ncv2.putText(\nim,\nrecords[-1][start:end], (0, int(40 * text_scale)),\n- cv2.FONT_HERSHEY_PLAIN,\n+ cv2.FONT_ITALIC,\ntext_scale, (0, 0, 255),\n- thickness=2)\n+ thickness=text_thickness)\nif num_classes == 1 and do_entrance_counting:\nentrance_line = tuple(map(int, entrance))\n@@ -229,9 +227,9 @@ def plot_tracking_dict(image,\ncv2.putText(\nim,\nrecords[-1][start:-1], (0, int(60 * text_scale)),\n- cv2.FONT_HERSHEY_PLAIN,\n+ cv2.FONT_ITALIC,\ntext_scale, (0, 0, 255),\n- thickness=2)\n+ thickness=text_thickness)\nfor cls_id in range(num_classes):\ntlwhs = tlwhs_dict[cls_id]\n@@ -240,10 +238,10 @@ def plot_tracking_dict(image,\ncv2.putText(\nim,\n'frame: %d fps: %.2f num: %d' % (frame_id, fps, len(tlwhs)),\n- (0, int(15 * text_scale)),\n- cv2.FONT_HERSHEY_PLAIN,\n+ (0, int(15 * text_scale) + 5),\n+ cv2.FONT_ITALIC,\ntext_scale, (0, 0, 255),\n- thickness=2)\n+ thickness=text_thickness)\nrecord_id = set()\nfor i, tlwh in enumerate(tlwhs):\n@@ -273,18 +271,18 @@ def plot_tracking_dict(image,\nthickness=line_thickness)\ncv2.putText(\nim,\n- id_text, (intbox[0], intbox[1] - 10),\n- cv2.FONT_HERSHEY_PLAIN,\n- text_scale, (0, 0, 255),\n+ id_text, (intbox[0], intbox[1] - 25),\n+ cv2.FONT_ITALIC,\n+ text_scale, (0, 255, 255),\nthickness=text_thickness)\nif scores is not None:\n- text = '{:.2f}'.format(float(scores[i]))\n+ text = 'score: {:.2f}'.format(float(scores[i]))\ncv2.putText(\nim,\n- text, (intbox[0], intbox[1] + 10),\n- cv2.FONT_HERSHEY_PLAIN,\n- text_scale, (0, 255, 255),\n+ text, (intbox[0], intbox[1] - 6),\n+ cv2.FONT_ITALIC,\n+ text_scale, (0, 255, 0),\nthickness=text_thickness)\nif center_traj is not None:\nfor traj in center_traj:\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/visualize.py",
"new_path": "deploy/python/visualize.py",
"diff": "@@ -338,17 +338,17 @@ def visualize_attr(im, results, boxes=None):\nim = np.ascontiguousarray(np.copy(im))\nim_h, im_w = im.shape[:2]\n- text_scale = max(1, int(im.shape[0] / 1600.))\n- text_thickness = 2\n+ text_scale = max(0.5, im.shape[0] / 3000.)\n+ text_thickness = 1\n- line_inter = im.shape[0] / 50.\n+ line_inter = im.shape[0] / 40.\nfor i, res in enumerate(results):\nif boxes is None:\n- text_w = 1\n+ text_w = 3\ntext_h = 1\nelse:\nbox = boxes[i]\n- text_w = int(box[2])\n+ text_w = int(box[2]) + 3\ntext_h = int(box[3])\nfor text in res:\ntext_h += int(line_inter)\n@@ -357,8 +357,8 @@ def visualize_attr(im, results, boxes=None):\nim,\ntext,\ntext_loc,\n- cv2.FONT_HERSHEY_PLAIN,\n- text_scale, (0, 0, 255),\n+ cv2.FONT_ITALIC,\n+ text_scale, (0, 255, 255),\nthickness=text_thickness)\nreturn im\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
refine pphuman vis (#5426)
|
499,395 |
22.03.2022 21:09:30
| -28,800 |
082b1ac9b23d5b21d45f77d707ef87d038e737cf
|
fix transform problem of RandomCrop
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/transform/operators.py",
"new_path": "ppdet/data/transform/operators.py",
"diff": "@@ -1504,6 +1504,11 @@ class RandomCrop(BaseOperator):\nif 'is_crowd' in sample:\nsample['is_crowd'] = np.take(\nsample['is_crowd'], valid_ids, axis=0)\n+\n+ if 'difficult' in sample:\n+ sample['difficult'] = np.take(\n+ sample['difficult'], valid_ids, axis=0)\n+\nreturn sample\nreturn sample\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix transform problem of RandomCrop (#5412)
|
499,398 |
23.03.2022 11:14:44
| -28,800 |
14e722043da22c4a43595ee3b415b2fe0197893d
|
[NPU] fix fp16
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -373,7 +373,7 @@ class Trainer(object):\n# enabel auto mixed precision mode\nif self.cfg.get('amp', False):\nscaler = amp.GradScaler(\n- enable=self.cfg.use_gpu, init_loss_scaling=1024)\n+ enable=self.cfg.use_gpu or self.cfg.use_npu, init_loss_scaling=1024)\nself.status.update({\n'epoch_id': self.start_epoch,\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[NPU] fix fp16 (#5417)
|
499,333 |
23.03.2022 15:31:40
| -28,800 |
ef342535e0eb4da49e401b5c6f4a9ff1af42fa6b
|
fix mask head when mask num is 1
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/mask_head.py",
"new_path": "ppdet/modeling/heads/mask_head.py",
"diff": "@@ -226,6 +226,10 @@ class MaskHead(nn.Layer):\nnum_masks = paddle.shape(mask_logit)[0]\nindex = paddle.arange(num_masks).cast('int32')\nmask_out = mask_logit[index, labels]\n+ mask_out_shape = paddle.shape(mask_out)\n+ mask_out = paddle.reshape(mask_out, [\n+ paddle.shape(index), mask_out_shape[-2], mask_out_shape[-1]\n+ ])\nmask_out = F.sigmoid(mask_out)\nreturn mask_out\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix mask head when mask num is 1 (#5436)
|
499,333 |
24.03.2022 14:12:32
| -28,800 |
5815fbc9faab8e9eff1c7004954ed698f38c5050
|
add message for export_onnx
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -553,6 +553,10 @@ class PredictConfig():\nself.nms = yml_conf['NMS']\nif 'fpn_stride' in yml_conf:\nself.fpn_stride = yml_conf['fpn_stride']\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+ )\nself.print_config()\ndef check_model(self, yml_conf):\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/export_utils.py",
"new_path": "ppdet/engine/export_utils.py",
"diff": "@@ -126,7 +126,13 @@ def _dump_infer_config(config, path, image_shape, model):\n'metric': config['metric'],\n'use_dynamic_shape': use_dynamic_shape\n})\n+ export_onnx = config.get('export_onnx', False)\n+\ninfer_arch = config['architecture']\n+ if 'RCNN' in infer_arch and export_onnx:\n+ logger.warning(\n+ \"Exporting RCNN model to ONNX only support batch_size = 1\")\n+ infer_cfg['export_onnx'] = True\nif infer_arch in MOT_ARCH:\nif infer_arch == 'DeepSORT':\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add message for export_onnx (#5444)
|
499,339 |
24.03.2022 16:12:18
| -28,800 |
d6ffa2b5e9f4d8725a8ecb168ea43431534268ae
|
[PPYOLOE] fix oom eval in train
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/optimizer_300e.yml",
"new_path": "configs/ppyoloe/_base_/optimizer_300e.yml",
"diff": "epoch: 300\nLearningRate:\n- base_lr: 0.03\n+ base_lr: 0.025\nschedulers:\n- !CosineDecay\nmax_epochs: 360\n- !LinearWarmup\n- start_factor: 0.001\n- steps: 3000\n+ start_factor: 0.\n+ epochs: 5\nOptimizerBuilder:\noptimizer:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_reader.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_reader.yml",
"diff": "-worker_num: 8\n+worker_num: 4\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -11,7 +11,7 @@ TrainReader:\n- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}\n- Permute: {}\n- PadGT: {}\n- batch_size: 24\n+ batch_size: 20\nshuffle: true\ndrop_last: true\nuse_shared_memory: true\n@@ -23,7 +23,7 @@ EvalReader:\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+ batch_size: 2\nTestReader:\ninputs_def:\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml",
"diff": "@@ -9,7 +9,6 @@ _BASE_: [\nlog_iter: 100\nsnapshot_epoch: 10\nweights: output/ppyoloe_crn_l_300e_coco/model_final\n-find_unused_parameters: True\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/CSPResNetb_l_pretrained.pdparams\ndepth_mult: 1.0\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/ppyoloe_crn_m_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_m_300e_coco.yml",
"diff": "@@ -9,20 +9,13 @@ _BASE_: [\nlog_iter: 100\nsnapshot_epoch: 10\nweights: output/ppyoloe_crn_m_300e_coco/model_final\n-find_unused_parameters: True\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/CSPResNetb_m_pretrained.pdparams\ndepth_mult: 0.67\nwidth_mult: 0.75\nTrainReader:\n- batch_size: 32\n+ batch_size: 28\nLearningRate:\n- base_lr: 0.04\n- schedulers:\n- - !CosineDecay\n- max_epochs: 360\n- - !LinearWarmup\n- start_factor: 0.001\n- steps: 2300\n+ base_lr: 0.035\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_s_300e_coco.yml",
"diff": "@@ -9,7 +9,6 @@ _BASE_: [\nlog_iter: 100\nsnapshot_epoch: 10\nweights: output/ppyoloe_crn_s_300e_coco/model_final\n-find_unused_parameters: True\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/CSPResNetb_s_pretrained.pdparams\ndepth_mult: 0.33\n@@ -20,9 +19,3 @@ TrainReader:\nLearningRate:\nbase_lr: 0.04\n- schedulers:\n- - !CosineDecay\n- max_epochs: 360\n- - !LinearWarmup\n- start_factor: 0.001\n- steps: 2300\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/ppyoloe_crn_x_300e_coco.yml",
"new_path": "configs/ppyoloe/ppyoloe_crn_x_300e_coco.yml",
"diff": "@@ -9,7 +9,6 @@ _BASE_: [\nlog_iter: 100\nsnapshot_epoch: 10\nweights: output/ppyoloe_crn_x_300e_coco/model_final\n-find_unused_parameters: True\npretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/CSPResNetb_x_pretrained.pdparams\ndepth_mult: 1.33\n@@ -20,9 +19,3 @@ TrainReader:\nLearningRate:\nbase_lr: 0.02\n- schedulers:\n- - !CosineDecay\n- max_epochs: 360\n- - !LinearWarmup\n- start_factor: 0.001\n- steps: 4600\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/atss_assigner.py",
"new_path": "ppdet/modeling/assigners/atss_assigner.py",
"diff": "@@ -183,9 +183,6 @@ class ATSSAssigner(nn.Layer):\nmask_positive)\nmask_positive_sum = mask_positive.sum(axis=-2)\nassigned_gt_index = mask_positive.argmax(axis=-2)\n- assert mask_positive_sum.max() == 1, \\\n- (\"one anchor just assign one gt, but received not equals 1. \"\n- \"Received: %f\" % mask_positive_sum.max().item())\n# assigned target\nbatch_ind = paddle.arange(\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"new_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"diff": "@@ -129,9 +129,6 @@ class TaskAlignedAssigner(nn.Layer):\nmask_positive)\nmask_positive_sum = mask_positive.sum(axis=-2)\nassigned_gt_index = mask_positive.argmax(axis=-2)\n- assert mask_positive_sum.max() == 1, \\\n- (\"one anchor just assign one gt, but received not equals 1. \"\n- \"Received: %f\" % mask_positive_sum.max().item())\n# assigned target\nassigned_gt_index = assigned_gt_index + batch_ind * num_max_boxes\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[PPYOLOE] fix oom eval in train (#5409)
|
499,348 |
25.03.2022 15:38:40
| -28,800 |
87b9d564ad761fe7db7a8ca47a676ecac6f7c3df
|
fix datacollector error without reid;test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/datacollector.py",
"new_path": "deploy/pphuman/datacollector.py",
"diff": "@@ -75,7 +75,8 @@ class DataCollector(object):\naction_res = Result.get('action')\nreid_res = Result.get('reid')\n- for idx, mot_item in enumerate(reid_res['rects']):\n+ rects = reid_res['rects'] if reid_res is not None else mot_res['boxes']\n+ for idx, mot_item in enumerate(rects):\nids = int(mot_item[0])\nif ids not in self.collector:\nself.collector[ids] = copy.deepcopy(self.mots)\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -460,6 +460,7 @@ class PipePredictor(object):\n# nothing detected\nif len(mot_res['boxes']) == 0:\nframe_id += 1\n+ if frame_id > self.warmup_frame:\nself.pipe_timer.img_num += 1\nself.pipe_timer.total_time.end()\nif self.cfg['visual']:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix datacollector error without reid;test=document_fix (#5459)
|
499,348 |
25.03.2022 17:36:30
| -28,800 |
abb60be9c9206990abafc5e2e3d46d2791a4c68a
|
add mtmct docs;
test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/datacollector.py",
"new_path": "deploy/pphuman/datacollector.py",
"diff": "@@ -86,7 +86,7 @@ class DataCollector(object):\nif attr_res:\nself.collector[ids][\"attrs\"].append(attr_res['output'][idx])\nif kpt_res:\n- self.collector[ids][\"kpts\"].append(kpt_res['output'][idx])\n+ self.collector[ids][\"kpts\"].append(kpt_res['keypoint'][idx])\nif action_res:\nself.collector[ids][\"actions\"].append(action_res['output'][idx])\nelse:\n"
},
{
"change_type": "ADD",
"old_path": "deploy/pphuman/docs/images/c1.gif",
"new_path": "deploy/pphuman/docs/images/c1.gif",
"diff": "Binary files /dev/null and b/deploy/pphuman/docs/images/c1.gif differ\n"
},
{
"change_type": "ADD",
"old_path": "deploy/pphuman/docs/images/c2.gif",
"new_path": "deploy/pphuman/docs/images/c2.gif",
"diff": "Binary files /dev/null and b/deploy/pphuman/docs/images/c2.gif differ\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add mtmct docs; (#5458)
test=document_fix
|
499,348 |
26.03.2022 14:39:06
| -28,800 |
49379331ad6dbb732c2942492bd66f8bd92728fb
|
update kpt and action data collector;
test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/datacollector.py",
"new_path": "deploy/pphuman/datacollector.py",
"diff": "@@ -86,9 +86,10 @@ class DataCollector(object):\nif attr_res:\nself.collector[ids][\"attrs\"].append(attr_res['output'][idx])\nif kpt_res:\n- self.collector[ids][\"kpts\"].append(kpt_res['keypoint'][idx])\n+ self.collector[ids][\"kpts\"].append(\n+ [kpt_res['keypoint'][0][idx], kpt_res['keypoint'][1][idx]])\nif action_res:\n- self.collector[ids][\"actions\"].append(action_res['output'][idx])\n+ self.collector[ids][\"actions\"].append(action_res[idx + 1])\nelse:\n# action model generate result per X frames, Not available every frames\nself.collector[ids][\"actions\"].append(None)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update kpt and action data collector; (#5464)
test=document_fix
|
499,348 |
27.03.2022 18:07:09
| -28,800 |
9631d37de4b627fbbf18b7d47def8473140444ce
|
fix print logs;
test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/mtmct.py",
"new_path": "deploy/pphuman/mtmct.py",
"diff": "@@ -111,7 +111,8 @@ def save_mtmct_vis_results(camera_results, captures, output_dir):\nfor idx, video_file in enumerate(captures):\ncapture = cv2.VideoCapture(video_file)\ncid = camera_ids[idx]\n- video_out_name = \"mtmct_vis_c\" + str(cid) + \".mp4\"\n+ basename = os.path.basename(video_file)\n+ video_out_name = \"vis_\" + basename\nprint(\"Start visualizing output video: {}\".format(video_out_name))\nout_path = os.path.join(save_dir, video_out_name)\n@@ -182,7 +183,7 @@ def get_dist_mat(x, y, func_name=\"euclidean\"):\ndist_mat = get_cosine(x, y)\nelif func_name == \"euclidean\":\ndist_mat = get_euclidean(x, y)\n- print(\"Using {func_name} as distance function during evaluation\")\n+ print(\"Using {} as distance function during evaluation\".format(func_name))\nreturn dist_mat\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix print logs; (#5466)
test=document_fix
|
499,333 |
28.03.2022 15:23:48
| -28,800 |
12fb0052564c5de546cd8071ba249cf2c4937e13
|
update data in pphuman, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pptracking/python/mot/visualize.py",
"new_path": "deploy/pptracking/python/mot/visualize.py",
"diff": "@@ -209,7 +209,7 @@ def plot_tracking_dict(image,\nend = records[-1].find('In')\ncv2.putText(\nim,\n- records[-1][start:end], (0, int(40 * text_scale)),\n+ records[-1][start:end], (0, int(40 * text_scale) + 10),\ncv2.FONT_ITALIC,\ntext_scale, (0, 0, 255),\nthickness=text_thickness)\n@@ -226,7 +226,7 @@ def plot_tracking_dict(image,\nstart = records[-1].find('In')\ncv2.putText(\nim,\n- records[-1][start:-1], (0, int(60 * text_scale)),\n+ records[-1][start:-1], (0, int(60 * text_scale) + 10),\ncv2.FONT_ITALIC,\ntext_scale, (0, 0, 255),\nthickness=text_thickness)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update data in pphuman, test=document_fix (#5474)
|
499,301 |
28.03.2022 20:00:44
| -28,800 |
2ef5f0cd218f7b29fc85b946a31db30ae0fcae72
|
update yoloe configs
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/README.md",
"new_path": "configs/ppyoloe/README.md",
"diff": "@@ -75,6 +75,12 @@ CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/ppyoloe/ppyoloe_crn_l_30\n### 4. Deployment\n+- PaddleInference [Python](../../deploy/python) & [C++](../../deploy/cpp)\n+- [Paddle-TensorRT](../../deploy/TENSOR_RT.md)\n+- [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX)\n+- [PaddleServing](https://github.com/PaddlePaddle/Serving)\n+<!-- - [Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite) -->\n+\nFor deployment on GPU or benchmarked, model should be first exported to inference model using `tools/export_model.py`.\nExporting PP-YOLOE for Paddle Inference **without TensorRT**, use following command.\n@@ -116,6 +122,20 @@ paddle2onnx --model_dir output_inference/ppyoloe_crn_l_300e_coco --model_filenam\n```\n+### 5. Other Datasets\n+\n+Model | AP | AP<sub>50</sub>\n+---|---|---\n+[YOLOX](https://github.com/Megvii-BaseDetection/YOLOX) | 22.6 | 37.5\n+[YOLOv5](https://github.com/ultralytics/yolov5) | 26.0 | 42.7\n+**PP-YOLOE** | **30.5** | **46.4**\n+\n+**Note**\n+- Here, we use [VisDrone](https://github.com/VisDrone/VisDrone-Dataset) dataset, and to detect 9 objects including `person, bicycles, car, van, truck, tricyle, awning-tricyle, bus, motor`.\n+- Above models trained using official default config, and load pretrained parameters on COCO dataset.\n+- *Due to the limited time, more verification results will be supplemented in the future. You are also welcome to contribute to PP-YOLOE*\n+\n+\n## Appendix\nAblation experiments of PP-YOLOE.\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update yoloe configs (#5478)
|
499,339 |
28.03.2022 20:01:25
| -28,800 |
839746dd865b5734461e90ce4d1aba2ea426a633
|
[TIPC] fix hrnet epoch num
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/keypoint/higherhrnet_hrnet_w32_512_train_infer_python.txt",
"new_path": "test_tipc/configs/keypoint/higherhrnet_hrnet_w32_512_train_infer_python.txt",
"diff": "@@ -52,7 +52,7 @@ inference:./deploy/python/keypoint_infer.py\n===========================train_benchmark_params==========================\nbatch_size:20|24\nfp_items:fp32|fp16\n-epoch:1\n+epoch:20\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\nflags:null\n===========================infer_benchmark_params===========================\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/configs/keypoint/hrnet_w32_256x192_train_infer_python.txt",
"new_path": "test_tipc/configs/keypoint/hrnet_w32_256x192_train_infer_python.txt",
"diff": "@@ -52,7 +52,7 @@ null:null\n===========================train_benchmark_params==========================\nbatch_size:64|160\nfp_items:fp32|fp16\n-epoch:1\n+epoch:40\n--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile\nflags:null\n===========================infer_benchmark_params===========================\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix hrnet epoch num (#5481)
|
499,395 |
28.03.2022 20:59:00
| -28,800 |
f3f8dc9cbdf6b0b7ba3fc758b692233658e413a6
|
fix ema resume
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer.py",
"diff": "@@ -339,6 +339,7 @@ class ModelEMA(object):\ndef resume(self, state_dict, step=0):\nfor k, v in state_dict.items():\n+ if k in self.state_dict:\nself.state_dict[k] = v\nself.step = step\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix ema resume (#5413)
|
499,395 |
28.03.2022 20:59:12
| -28,800 |
504e89ba580112f481ff9d8698ea4f7c234ba745
|
fix ppyoloe docs, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/README.md",
"new_path": "configs/ppyoloe/README.md",
"diff": "@@ -92,7 +92,7 @@ python tools/export_model.py configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o weig\nExporting PP-YOLOE for Paddle Inference **with TensorRT** for better performance, use following command with extra `-o trt=True` setting.\n```bash\n-python tools/export_model.py configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams -o trt=True\n+python tools/export_model.py configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams trt=True\n```\n`deploy/python/infer.py` is used to load exported paddle inference model above for inference and benchmark through PaddleInference.\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix ppyoloe docs, test=document_fix (#5484)
|
499,339 |
28.03.2022 21:39:43
| -28,800 |
98aebc4603640ab8f603bad696e17ad41e8e57e8
|
refine coslr, add last_plateau_epochs
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer.py",
"diff": "@@ -41,12 +41,21 @@ class CosineDecay(object):\nmax_epochs (int): max epochs for the training process.\nif you commbine cosine decay with warmup, it is recommended that\nthe max_iters is much larger than the warmup iter\n+ use_warmup (bool): whether to use warmup. Default: True.\n+ min_lr_ratio (float): minimum learning rate ratio. Default: 0.\n+ last_plateau_epochs (int): use minimum learning rate in\n+ the last few epochs. Default: 0.\n\"\"\"\n- def __init__(self, max_epochs=1000, use_warmup=True, eta_min=0.):\n+ def __init__(self,\n+ max_epochs=1000,\n+ use_warmup=True,\n+ min_lr_ratio=0.,\n+ last_plateau_epochs=0):\nself.max_epochs = max_epochs\nself.use_warmup = use_warmup\n- self.eta_min = eta_min\n+ self.min_lr_ratio = min_lr_ratio\n+ self.last_plateau_epochs = last_plateau_epochs\ndef __call__(self,\nbase_lr=None,\n@@ -56,21 +65,38 @@ class CosineDecay(object):\nassert base_lr is not None, \"either base LR or values should be provided\"\nmax_iters = self.max_epochs * int(step_per_epoch)\n-\n+ last_plateau_iters = self.last_plateau_epochs * int(step_per_epoch)\n+ min_lr = base_lr * self.min_lr_ratio\nif boundary is not None and value is not None and self.use_warmup:\n+ # use warmup\nwarmup_iters = len(boundary)\nfor i in range(int(boundary[-1]), max_iters):\nboundary.append(i)\n-\n- decayed_lr = base_lr * 0.5 * (math.cos(\n+ if i < max_iters - last_plateau_iters:\n+ decayed_lr = min_lr + (base_lr - min_lr) * 0.5 * (math.cos(\n(i - warmup_iters) * math.pi /\n- (max_iters - warmup_iters)) + 1)\n- decayed_lr = decayed_lr if decayed_lr > self.eta_min else self.eta_min\n+ (max_iters - warmup_iters - last_plateau_iters)) + 1)\nvalue.append(decayed_lr)\n+ else:\n+ value.append(min_lr)\n+ return optimizer.lr.PiecewiseDecay(boundary, value)\n+ elif last_plateau_iters > 0:\n+ # not use warmup, but set `last_plateau_epochs` > 0\n+ boundary = []\n+ value = []\n+ for i in range(max_iters):\n+ if i < max_iters - last_plateau_iters:\n+ decayed_lr = min_lr + (base_lr - min_lr) * 0.5 * (math.cos(\n+ i * math.pi / (max_iters - last_plateau_iters)) + 1)\n+ value.append(decayed_lr)\n+ else:\n+ value.append(min_lr)\n+ if i > 0:\n+ boundary.append(i)\nreturn optimizer.lr.PiecewiseDecay(boundary, value)\nreturn optimizer.lr.CosineAnnealingDecay(\n- base_lr, T_max=max_iters, eta_min=self.eta_min)\n+ base_lr, T_max=max_iters, eta_min=min_lr)\n@serializable\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
refine coslr, add last_plateau_epochs (#5401)
|
499,348 |
28.03.2022 22:14:31
| -28,800 |
cd08a2450f71a9d657e600bf4294e5196b24fdae
|
fix some problem in lite deploy
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/lite/include/config_parser.h",
"new_path": "deploy/lite/include/config_parser.h",
"diff": "namespace PaddleDetection {\n-void load_jsonf(std::string jsonfile, Json::Value& jsondata);\n+void load_jsonf(std::string jsonfile, const Json::Value& jsondata);\n// Inference model configuration parser\nclass ConfigPaser {\n@@ -43,12 +43,13 @@ class ConfigPaser {\nJson::Value config;\nload_jsonf(model_dir + OS_PATH_SEP + cfg + \".json\", config);\n- // Get model arch : YOLO, SSD, RetinaNet, RCNN, Face\n+ // Get model arch : YOLO, SSD, RetinaNet, RCNN, Face, PicoDet, HRNet\nif (config.isMember(\"arch\")) {\narch_ = config[\"arch\"].as<std::string>();\n} else {\n- std::cerr << \"Please set model arch,\"\n- << \"support value : YOLO, SSD, RetinaNet, RCNN, Face.\"\n+ std::cerr\n+ << \"Please set model arch,\"\n+ << \"support value : YOLO, SSD, RetinaNet, RCNN, Face, PicoDet, HRNet.\"\n<< std::endl;\nreturn false;\n}\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/lite/src/config_parser.cc",
"new_path": "deploy/lite/src/config_parser.cc",
"diff": "namespace PaddleDetection {\n-void load_jsonf(std::string jsonfile, Json::Value &jsondata) {\n+void load_jsonf(std::string jsonfile, const Json::Value &jsondata) {\nstd::ifstream ifs;\nifs.open(jsonfile);\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/lite/src/main.cc",
"new_path": "deploy/lite/src/main.cc",
"diff": "@@ -43,10 +43,8 @@ void PrintBenchmarkLog(std::vector<double> det_time, int img_num) {\n<< std::endl;\nRT_Config[\"model_dir_det\"].as<std::string>().erase(\nRT_Config[\"model_dir_det\"].as<std::string>().find_last_not_of(\"/\") + 1);\n- std::cout\n- << \"detection model_name: \"\n- << RT_Config[\"model_dir_det\"].as<std::string>()\n- << std::endl;\n+ std::cout << \"detection model_name: \"\n+ << RT_Config[\"model_dir_det\"].as<std::string>() << std::endl;\nstd::cout << \"----------------------- Perf info ------------------------\"\n<< std::endl;\nstd::cout << \"Total number of predicted data: \" << img_num\n@@ -69,14 +67,14 @@ void PrintKptsBenchmarkLog(std::vector<double> det_time, int img_num){\nRT_Config[\"model_dir_keypoint\"].as<std::string>().erase(\nRT_Config[\"model_dir_keypoint\"].as<std::string>().find_last_not_of(\"/\") +\n1);\n- std::cout\n- << \"keypoint model_name: \"\n+ std::cout << \"keypoint model_name: \"\n<< RT_Config[\"model_dir_keypoint\"].as<std::string>() << std::endl;\nstd::cout << \"----------------------- Perf info ------------------------\"\n<< std::endl;\nstd::cout << \"Total number of predicted data: \" << img_num\n<< \" and total time spent(ms): \"\n- << std::accumulate(det_time.begin(), det_time.end(), 0.) << std::endl;\n+ << std::accumulate(det_time.begin(), det_time.end(), 0.)\n+ << std::endl;\nimg_num = std::max(1, img_num);\nstd::cout << \"Average time cost per person:\" << std::endl\n<< \"preproce_time(ms): \" << det_time[0] / img_num\n@@ -136,7 +134,7 @@ void PredictImage(const std::vector<std::string> all_img_paths,\nPaddleDetection::KeyPointDetector* keypoint,\nconst std::string& output_dir = \"output\") {\nstd::vector<double> det_t = {0, 0, 0};\n- int steps = ceil(float(all_img_paths.size()) / batch_size_det);\n+ int steps = ceil(static_cast<float>(all_img_paths.size()) / batch_size_det);\nint kpts_imgs = 0;\nstd::vector<double> keypoint_t = {0, 0, 0};\ndouble midtimecost = 0;\n@@ -243,7 +241,7 @@ void PredictImage(const std::vector<std::string> all_img_paths,\nstd::chrono::duration<float> midtimediff =\nkeypoint_crop_time - keypoint_start_time;\n- midtimecost += double(midtimediff.count() * 1000);\n+ midtimecost += static_cast<double>(midtimediff.count() * 1000);\nif (imgs_kpts.size() == RT_Config[\"batch_size_keypoint\"].as<int>() ||\n((i == imsize - 1) && !imgs_kpts.empty())) {\n@@ -275,8 +273,8 @@ void PredictImage(const std::vector<std::string> all_img_paths,\nstd::string kpts_savepath =\noutput_path + \"keypoint_\" +\nimage_file_path.substr(image_file_path.find_last_of('/') + 1);\n- cv::Mat kpts_vis_img =\n- VisualizeKptsResult(im, result_kpts, colormap_kpts, keypoint->get_threshold());\n+ cv::Mat kpts_vis_img = VisualizeKptsResult(\n+ im, result_kpts, colormap_kpts, keypoint->get_threshold());\ncv::imwrite(kpts_savepath, kpts_vis_img, compression_params);\nprintf(\"Visualized output saved as %s\\n\", kpts_savepath.c_str());\n} else {\n@@ -298,16 +296,15 @@ void PredictImage(const std::vector<std::string> all_img_paths,\nPrintBenchmarkLog(det_t, all_img_paths.size());\nif (keypoint) {\nPrintKptsBenchmarkLog(keypoint_t, kpts_imgs);\n- PrintTotalIimeLog((det_t[0] + det_t[1] + det_t[2]) / all_img_paths.size(),\n+ PrintTotalIimeLog(\n+ (det_t[0] + det_t[1] + det_t[2]) / all_img_paths.size(),\n(keypoint_t[0] + keypoint_t[1] + keypoint_t[2]) / all_img_paths.size(),\nmidtimecost / all_img_paths.size());\n}\n-\n}\nint main(int argc, char** argv) {\n- std::cout << \"Usage: \" << argv[0]\n- << \" [config_path](option) [image_dir](option)\\n\";\n+ std::cout << \"Usage: \" << argv[0] << \" [config_path] [image_dir](option)\\n\";\nif (argc < 2) {\nstd::cout << \"Usage: ./main det_runtime_config.json\" << std::endl;\nreturn -1;\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix some problem in lite deploy (#5488)
|
499,385 |
29.03.2022 14:53:39
| -28,800 |
f4b8dded1ef7555325997cedd056fdb8f728ac9d
|
Update docs for PP-YOLOE
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/README.md",
"new_path": "configs/ppyoloe/README.md",
"diff": "@@ -86,13 +86,13 @@ For deployment on GPU or benchmarked, model should be first exported to inferenc\nExporting PP-YOLOE for Paddle Inference **without TensorRT**, use following command.\n```bash\n-python tools/export_model.py configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams\n+python tools/export_model.py -c configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams\n```\nExporting PP-YOLOE for Paddle Inference **with TensorRT** for better performance, use following command with extra `-o trt=True` setting.\n```bash\n-python tools/export_model.py configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams trt=True\n+python tools/export_model.py -c configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams trt=True\n```\n`deploy/python/infer.py` is used to load exported paddle inference model above for inference and benchmark through Paddle Inference.\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
Update docs for PP-YOLOE (#5494)
|
499,333 |
29.03.2022 21:11:25
| -28,800 |
0743e380e47def3e45a2e6f59067903958e3d285
|
fix test arch
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/tests/test_architectures.py",
"new_path": "ppdet/modeling/tests/test_architectures.py",
"diff": "@@ -62,7 +62,7 @@ class TestGFL(TestFasterRCNN):\nclass TestPicoDet(TestFasterRCNN):\ndef set_config(self):\n- self.cfg_file = 'configs/picodet/picodet_s_320_coco.yml'\n+ self.cfg_file = 'configs/picodet/picodet_s_320_coco_lcnet.yml'\nif __name__ == '__main__':\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix test arch (#5503)
|
499,395 |
30.03.2022 15:41:32
| -28,800 |
8d3b8b47838e7c73fab59117288c52233e70b812
|
modify fuse normalize
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/architectures/meta_arch.py",
"new_path": "ppdet/modeling/architectures/meta_arch.py",
"diff": "@@ -22,22 +22,23 @@ class BaseArch(nn.Layer):\nself.fuse_norm = False\ndef load_meanstd(self, cfg_transform):\n- self.scale = 1.\n- self.mean = paddle.to_tensor([0.485, 0.456, 0.406]).reshape(\n- (1, 3, 1, 1))\n- self.std = paddle.to_tensor([0.229, 0.224, 0.225]).reshape((1, 3, 1, 1))\n+ scale = 1.\n+ mean = np.array([0.485, 0.456, 0.406], dtype=np.float32)\n+ std = np.array([0.229, 0.224, 0.225], dtype=np.float32)\nfor item in cfg_transform:\nif 'NormalizeImage' in item:\n- self.mean = paddle.to_tensor(item['NormalizeImage'][\n- 'mean']).reshape((1, 3, 1, 1))\n- self.std = paddle.to_tensor(item['NormalizeImage'][\n- 'std']).reshape((1, 3, 1, 1))\n+ mean = np.array(\n+ item['NormalizeImage']['mean'], dtype=np.float32)\n+ std = np.array(item['NormalizeImage']['std'], dtype=np.float32)\nif item['NormalizeImage'].get('is_scale', True):\n- self.scale = 1. / 255.\n+ scale = 1. / 255.\nbreak\nif self.data_format == 'NHWC':\n- self.mean = self.mean.reshape(1, 1, 1, 3)\n- self.std = self.std.reshape(1, 1, 1, 3)\n+ self.scale = paddle.to_tensor(scale / std).reshape((1, 1, 1, 3))\n+ self.bias = paddle.to_tensor(-mean / std).reshape((1, 1, 1, 3))\n+ else:\n+ self.scale = paddle.to_tensor(scale / std).reshape((1, 3, 1, 1))\n+ self.bias = paddle.to_tensor(-mean / std).reshape((1, 3, 1, 1))\ndef forward(self, inputs):\nif self.data_format == 'NHWC':\n@@ -46,7 +47,7 @@ class BaseArch(nn.Layer):\nif self.fuse_norm:\nimage = inputs['image']\n- self.inputs['image'] = (image * self.scale - self.mean) / self.std\n+ self.inputs['image'] = image * self.scale + self.bias\nself.inputs['im_shape'] = inputs['im_shape']\nself.inputs['scale_factor'] = inputs['scale_factor']\nelse:\n@@ -66,8 +67,7 @@ class BaseArch(nn.Layer):\nouts = []\nfor inp in inputs_list:\nif self.fuse_norm:\n- self.inputs['image'] = (\n- inp['image'] * self.scale - self.mean) / self.std\n+ self.inputs['image'] = inp['image'] * self.scale + self.bias\nself.inputs['im_shape'] = inp['im_shape']\nself.inputs['scale_factor'] = inp['scale_factor']\nelse:\n@@ -92,7 +92,9 @@ class BaseArch(nn.Layer):\nkeep_top_k = self.bbox_post_process.nms.keep_top_k\nnms_threshold = self.bbox_post_process.nms.nms_threshold\nelse:\n- raise Exception(\"Multi scale test only supports CascadeRCNN, FasterRCNN and MaskRCNN for now\")\n+ raise Exception(\n+ \"Multi scale test only supports CascadeRCNN, FasterRCNN and MaskRCNN for now\"\n+ )\nfinal_boxes = []\nall_scale_outs = paddle.concat([o['bbox'] for o in outs]).numpy()\n@@ -101,9 +103,11 @@ class BaseArch(nn.Layer):\nif np.count_nonzero(idxs) == 0:\ncontinue\nr = nms(all_scale_outs[idxs, 1:], nms_threshold)\n- final_boxes.append(np.concatenate([np.full((r.shape[0], 1), c), r], 1))\n+ final_boxes.append(\n+ np.concatenate([np.full((r.shape[0], 1), c), r], 1))\nout = np.concatenate(final_boxes)\n- out = np.concatenate(sorted(out, key=lambda e: e[1])[-keep_top_k:]).reshape((-1, 6))\n+ out = np.concatenate(sorted(\n+ out, key=lambda e: e[1])[-keep_top_k:]).reshape((-1, 6))\nout = {\n'bbox': paddle.to_tensor(out),\n'bbox_num': paddle.to_tensor(np.array([out.shape[0], ]))\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
modify fuse normalize (#5513)
|
499,299 |
30.03.2022 17:37:29
| -28,800 |
35c9da7a61d18a0bd9892b4ac96fc86a0371ff5f
|
fix cpp inference trt error when bs>1
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/src/object_detector.cc",
"new_path": "deploy/cpp/src/object_detector.cc",
"diff": "// for setprecision\n#include <chrono>\n#include <iomanip>\n-#include \"include/object_detector.h\"\n-using namespace paddle_infer;\n+#include \"include/object_detector.h\"\nnamespace PaddleDetection {\n@@ -42,27 +41,22 @@ 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- // set DynamicShsape for image tensor\n+ // set DynamicShape for image tensor\nconst std::vector<int> min_input_shape = {\n- 1, 3, this->trt_min_shape_, this->trt_min_shape_};\n+ batch_size, 3, this->trt_min_shape_, this->trt_min_shape_};\nconst std::vector<int> max_input_shape = {\n- 1, 3, this->trt_max_shape_, this->trt_max_shape_};\n+ batch_size, 3, this->trt_max_shape_, this->trt_max_shape_};\nconst std::vector<int> opt_input_shape = {\n- 1, 3, this->trt_opt_shape_, this->trt_opt_shape_};\n+ batch_size, 3, this->trt_opt_shape_, this->trt_opt_shape_};\nconst std::map<std::string, std::vector<int>> map_min_input_shape = {\n{\"image\", min_input_shape}};\nconst std::map<std::string, std::vector<int>> map_max_input_shape = {\n@@ -70,8 +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@@ -96,12 +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();\nfor (int i = 0; i < results.size(); ++i) {\n// Configure color and text size\n@@ -142,19 +135,13 @@ 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@@ -169,8 +156,7 @@ 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> bbox_num, std::vector<float> output_data_,\nbool is_rbox = false) {\nresult->clear();\nint start_idx = 0;\n@@ -226,8 +212,7 @@ void ObjectDetector::Postprocess(\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@@ -255,9 +240,8 @@ 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- // TODO: reduce cost time\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@@ -278,8 +262,8 @@ 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@@ -320,8 +304,8 @@ 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(), 1,\n+ std::multiplies<int>());\nif (output_tensor->type() == paddle_infer::DataType::INT32) {\nout_bbox_num_data_.resize(out_num);\noutput_tensor->CopyToCpu(out_bbox_num_data_.data());\n@@ -344,8 +328,8 @@ 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(), 1,\n+ std::multiplies<int>());\noutput_shape_list.push_back(output_shape);\nif (output_tensor->type() == paddle_infer::DataType::INT32) {\nout_bbox_num_data_.resize(out_num);\n@@ -372,21 +356,14 @@ 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 {\nis_rbox = output_shape_list[0][output_shape_list[0].size() - 1] % 10 == 0;\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix cpp inference trt error when bs>1 (#5518)
|
499,339 |
30.03.2022 18:53:33
| -28,800 |
eb8e1541822997bad13270430de2b9f9509e3fee
|
[deploy] fix benchmark repeats
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/infer.py",
"new_path": "deploy/python/infer.py",
"diff": "@@ -231,7 +231,7 @@ class Detector(object):\nself.det_times.preprocess_time_s.end()\n# model prediction\n- result = self.predict(repeats=repeats) # warmup\n+ result = self.predict(repeats=50) # warmup\nself.det_times.inference_time_s.start()\nresult = self.predict(repeats=repeats)\nself.det_times.inference_time_s.end(repeats=repeats)\n@@ -790,7 +790,7 @@ 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- detector.predict_image(img_list, FLAGS.run_benchmark, repeats=10)\n+ detector.predict_image(img_list, FLAGS.run_benchmark, repeats=100)\nif not FLAGS.run_benchmark:\ndetector.det_times.info(average=True)\nelse:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[deploy] fix benchmark repeats (#5526)
|
499,339 |
30.03.2022 21:26:17
| -28,800 |
2cd40bdd8236bca312bb4594c45a733a5ddaf735
|
fix one_hot error
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/atss_assigner.py",
"new_path": "ppdet/modeling/assigners/atss_assigner.py",
"diff": "@@ -199,7 +199,11 @@ class ATSSAssigner(nn.Layer):\ngt_bboxes.reshape([-1, 4]), assigned_gt_index.flatten(), axis=0)\nassigned_bboxes = assigned_bboxes.reshape([batch_size, num_anchors, 4])\n- assigned_scores = F.one_hot(assigned_labels, self.num_classes)\n+ assigned_scores = F.one_hot(assigned_labels, self.num_classes + 1)\n+ ind = list(range(self.num_classes + 1))\n+ ind.remove(bg_index)\n+ assigned_scores = paddle.index_select(\n+ assigned_scores, paddle.to_tensor(ind), axis=-1)\nif pred_bboxes is not None:\n# assigned iou\nious = batch_iou_similarity(gt_bboxes, pred_bboxes) * mask_positive\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"new_path": "ppdet/modeling/assigners/task_aligned_assigner.py",
"diff": "@@ -143,7 +143,11 @@ class TaskAlignedAssigner(nn.Layer):\ngt_bboxes.reshape([-1, 4]), assigned_gt_index.flatten(), axis=0)\nassigned_bboxes = assigned_bboxes.reshape([batch_size, num_anchors, 4])\n- assigned_scores = F.one_hot(assigned_labels, num_classes)\n+ assigned_scores = F.one_hot(assigned_labels, num_classes + 1)\n+ ind = list(range(num_classes + 1))\n+ ind.remove(bg_index)\n+ assigned_scores = paddle.index_select(\n+ assigned_scores, paddle.to_tensor(ind), axis=-1)\n# rescale alignment metrics\nalignment_metrics *= mask_positive\nmax_metrics_per_instance = alignment_metrics.max(axis=-1, keepdim=True)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/ppyoloe_head.py",
"new_path": "ppdet/modeling/heads/ppyoloe_head.py",
"diff": "@@ -331,7 +331,8 @@ class PPYOLOEHead(nn.Layer):\nassigned_bboxes /= stride_tensor\n# cls loss\nif self.use_varifocal_loss:\n- one_hot_label = F.one_hot(assigned_labels, self.num_classes)\n+ one_hot_label = F.one_hot(assigned_labels,\n+ self.num_classes + 1)[..., :-1]\nloss_cls = self._varifocal_loss(pred_scores, assigned_scores,\none_hot_label)\nelse:\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/losses/detr_loss.py",
"new_path": "ppdet/modeling/losses/detr_loss.py",
"diff": "@@ -80,7 +80,7 @@ class DETRLoss(nn.Layer):\ntarget_label = target_label.reshape([bs, num_query_objects])\nif self.use_focal_loss:\ntarget_label = F.one_hot(target_label,\n- self.num_classes + 1)[:, :, :-1]\n+ self.num_classes + 1)[..., :-1]\nreturn {\n'loss_class': self.loss_coeff['class'] * sigmoid_focal_loss(\nlogits, target_label, num_gts / num_query_objects)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix one_hot error (#5532)
|
499,339 |
31.03.2022 13:33:20
| -28,800 |
b7dd8133b382a5e4500bf3a077acf8215e3fd536
|
[PPYOLOE] update doc, params, flops
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -373,7 +373,8 @@ class Trainer(object):\n# enabel auto mixed precision mode\nif self.cfg.get('amp', False):\nscaler = amp.GradScaler(\n- enable=self.cfg.use_gpu or self.cfg.use_npu, init_loss_scaling=1024)\n+ enable=self.cfg.use_gpu or self.cfg.use_npu,\n+ init_loss_scaling=1024)\nself.status.update({\n'epoch_id': self.start_epoch,\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/backbones/cspresnet.py",
"new_path": "ppdet/modeling/backbones/cspresnet.py",
"diff": "@@ -96,6 +96,8 @@ class RepVggBlock(nn.Layer):\nkernel, bias = self.get_equivalent_kernel_bias()\nself.conv.weight.set_value(kernel)\nself.conv.bias.set_value(bias)\n+ self.__delattr__('conv1')\n+ self.__delattr__('conv2')\ndef get_equivalent_kernel_bias(self):\nkernel3x3, bias3x3 = self._fuse_bn_tensor(self.conv1)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[PPYOLOE] update doc, params, flops (#5543)
|
499,333 |
31.03.2022 19:12:19
| -28,800 |
66e4f30a90eb98b777337177a6c35ee508297415
|
fix camera id in pphuman
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -400,7 +400,11 @@ class PipePredictor(object):\ncpu_threads, enable_mkldnn)\ndef set_file_name(self, path):\n+ if path is not None:\nself.file_name = os.path.split(path)[-1]\n+ else:\n+ # use camera id\n+ self.file_name = None\ndef get_result(self):\nreturn self.collector.get_res()\n@@ -533,6 +537,11 @@ class PipePredictor(object):\nim = self.visualize_video(frame, mot_res, frame_id,\nfps) # visualize\nwriter.write(im)\n+ if self.file_name is None: # use camera_id\n+ cv2.imshow('PPHuman', im)\n+ if cv2.waitKey(1) & 0xFF == ord('q'):\n+ break\n+\ncontinue\nself.pipeline_res.update(mot_res, 'mot')\n@@ -619,6 +628,10 @@ class PipePredictor(object):\nfps, entrance, records,\ncenter_traj) # visualize\nwriter.write(im)\n+ if self.file_name is None: # use camera_id\n+ cv2.imshow('PPHuman', im)\n+ if cv2.waitKey(1) & 0xFF == ord('q'):\n+ break\nwriter.release()\nprint('save result to {}'.format(out_path))\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix camera id in pphuman (#5553)
|
499,394 |
02.04.2022 11:18:34
| -28,800 |
18f8eaf8d9b71b0dfd7cf4713aec8843b0c1aed6
|
enable eval_mot.py to support use_xpu
|
[
{
"change_type": "MODIFY",
"old_path": "tools/eval_mot.py",
"new_path": "tools/eval_mot.py",
"diff": "@@ -108,8 +108,17 @@ def main():\ncheck_gpu(cfg.use_gpu)\ncheck_version()\n- place = 'gpu:{}'.format(ParallelEnv().dev_id) if cfg.use_gpu else 'cpu'\n- place = paddle.set_device(place)\n+ # disable xpu in config by default\n+ if 'use_xpu' not in cfg:\n+ cfg.use_xpu = False\n+\n+ if cfg.use_gpu:\n+ place = paddle.set_device('gpu')\n+ elif cfg.use_xpu:\n+ place = paddle.set_device('xpu')\n+ else:\n+ place = paddle.set_device('cpu')\n+\nrun(FLAGS, cfg)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
enable eval_mot.py to support use_xpu (#5570)
|
499,339 |
02.04.2022 21:14:27
| -28,800 |
d9b219a7ebf25ced8bb1bcd46bf11b6f287b3393
|
[PPYOLOE] alter eval_input_size
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_crn.yml",
"diff": "@@ -28,7 +28,6 @@ PPYOLOEHead:\ngrid_cell_offset: 0.5\nstatic_assigner_epoch: 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/ppyoloe/_base_/ppyoloe_reader.yml",
"new_path": "configs/ppyoloe/_base_/ppyoloe_reader.yml",
"diff": "worker_num: 4\n+eval_height: &eval_height 640\n+eval_width: &eval_width 640\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -20,17 +24,17 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}\n+ - Resize: {target_size: *eval_size, 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: {}\nbatch_size: 2\nTestReader:\ninputs_def:\n- image_shape: [3, 640, 640]\n+ image_shape: [3, *eval_height, *eval_width]\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}\n+ - Resize: {target_size: *eval_size, 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: {}\nbatch_size: 1\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/ppyoloe_head.py",
"new_path": "ppdet/modeling/heads/ppyoloe_head.py",
"diff": "@@ -45,7 +45,7 @@ class ESEAttn(nn.Layer):\n@register\nclass PPYOLOEHead(nn.Layer):\n- __shared__ = ['num_classes', 'trt', 'exclude_nms']\n+ __shared__ = ['num_classes', 'eval_size', 'trt', 'exclude_nms']\n__inject__ = ['static_assigner', 'assigner', 'nms']\ndef __init__(self,\n@@ -61,7 +61,7 @@ class PPYOLOEHead(nn.Layer):\nstatic_assigner='ATSSAssigner',\nassigner='TaskAlignedAssigner',\nnms='MultiClassNMS',\n- eval_input_size=[],\n+ eval_size=None,\nloss_weight={\n'class': 1.0,\n'iou': 2.5,\n@@ -80,7 +80,7 @@ class PPYOLOEHead(nn.Layer):\nself.iou_loss = GIoULoss()\nself.loss_weight = loss_weight\nself.use_varifocal_loss = use_varifocal_loss\n- self.eval_input_size = eval_input_size\n+ self.eval_size = eval_size\nself.static_assigner_epoch = static_assigner_epoch\nself.static_assigner = static_assigner\n@@ -127,7 +127,7 @@ class PPYOLOEHead(nn.Layer):\nself.proj.reshape([1, self.reg_max + 1, 1, 1]))\nself.proj_conv.weight.stop_gradient = True\n- if self.eval_input_size:\n+ if self.eval_size:\nanchor_points, stride_tensor = self._generate_anchors()\nself.register_buffer('anchor_points', anchor_points)\nself.register_buffer('stride_tensor', stride_tensor)\n@@ -164,8 +164,8 @@ class PPYOLOEHead(nn.Layer):\nif feats is not None:\n_, _, h, w = feats[i].shape\nelse:\n- h = int(self.eval_input_size[0] / stride)\n- w = int(self.eval_input_size[1] / stride)\n+ h = int(self.eval_size[0] / stride)\n+ w = int(self.eval_size[1] / stride)\nshift_x = paddle.arange(end=w) + self.grid_cell_offset\nshift_y = paddle.arange(end=h) + self.grid_cell_offset\nshift_y, shift_x = paddle.meshgrid(shift_y, shift_x)\n@@ -181,7 +181,7 @@ class PPYOLOEHead(nn.Layer):\nreturn anchor_points, stride_tensor\ndef forward_eval(self, feats):\n- if self.eval_input_size:\n+ if self.eval_size:\nanchor_points, stride_tensor = self.anchor_points, self.stride_tensor\nelse:\nanchor_points, stride_tensor = self._generate_anchors(feats)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[PPYOLOE] alter eval_input_size (#5569)
|
499,319 |
02.04.2022 22:32:30
| -28,800 |
28f791573ba12cb9954df26706973d2cabbbe141
|
[picodet]openvino benchmark
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/README_en.md",
"new_path": "configs/picodet/README_en.md",
"diff": "@@ -33,20 +33,20 @@ We developed a series of lightweight models, named `PP-PicoDet`. Because of the\n| Model | Input size | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params<br><sup>(M) | FLOPS<br><sup>(G) | Latency<sup><small>[CPU](#latency)</small><sup><br><sup>(ms) | Latency<sup><small>[Lite](#latency)</small><sup><br><sup>(ms) | Download | Config |\n| :-------- | :--------: | :---------------------: | :----------------: | :----------------: | :---------------: | :-----------------------------: | :-----------------------------: | :----------------------------------------: | :--------------------------------------- |\n-| PicoDet-XS | 320*320 | 23.5 | 36.1 | 0.70 | 0.67 | 10.9ms | 7.81ms | [model](https://paddledet.bj.bcebos.com/models/picodet_xs_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_xs_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_xs_320_coco_lcnet.yml) |\n-| PicoDet-XS | 416*416 | 26.2 | 39.3 | 0.70 | 1.13 | 15.4ms | 12.38ms | [model](https://paddledet.bj.bcebos.com/models/picodet_xs_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_xs_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_xs_416_coco_lcnet.yml) |\n-| PicoDet-S | 320*320 | 29.1 | 43.4 | 1.18 | 0.97 | 12.6ms | 9.56ms | [model](https://paddledet.bj.bcebos.com/models/picodet_s_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_s_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_s_320_coco_lcnet.yml) |\n-| PicoDet-S | 416*416 | 32.5 | 47.6 | 1.18 | 1.65 | 17.2ms | 15.20ms | [model](https://paddledet.bj.bcebos.com/models/picodet_s_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_s_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_s_416_coco_lcnet.yml) |\n-| PicoDet-M | 320*320 | 34.4 | 50.0 | 3.46 | 2.57 | 14.5ms | 17.68ms | [model](https://paddledet.bj.bcebos.com/models/picodet_m_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_m_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_m_320_coco_lcnet.yml) |\n-| PicoDet-M | 416*416 | 37.5 | 53.4 | 3.46 | 4.34 | 19.5ms | 28.39ms | [model](https://paddledet.bj.bcebos.com/models/picodet_m_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_m_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_m_416_coco_lcnet.yml) |\n-| PicoDet-L | 320*320 | 36.1 | 52.0 | 5.80 | 4.20 | 18.3ms | 25.21ms | [model](https://paddledet.bj.bcebos.com/models/picodet_l_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_l_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_l_320_coco_lcnet.yml) |\n-| PicoDet-L | 416*416 | 39.4 | 55.7 | 5.80 | 7.10 | 22.1ms | 42.23ms | [model](https://paddledet.bj.bcebos.com/models/picodet_l_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_l_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_l_416_coco_lcnet.yml) |\n-| PicoDet-L | 640*640 | 42.6 | 59.2 | 5.80 | 16.81 | 43.1ms | 108.1ms | [model](https://paddledet.bj.bcebos.com/models/picodet_l_640_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_l_640_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_l_640_coco_lcnet.yml) |\n+| PicoDet-XS | 320*320 | 23.5 | 36.1 | 0.70 | 0.67 | 3.9ms | 7.81ms | [model](https://paddledet.bj.bcebos.com/models/picodet_xs_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_xs_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_xs_320_coco_lcnet.yml) |\n+| PicoDet-XS | 416*416 | 26.2 | 39.3 | 0.70 | 1.13 | 6.1ms | 12.38ms | [model](https://paddledet.bj.bcebos.com/models/picodet_xs_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_xs_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_xs_416_coco_lcnet.yml) |\n+| PicoDet-S | 320*320 | 29.1 | 43.4 | 1.18 | 0.97 | 4.8ms | 9.56ms | [model](https://paddledet.bj.bcebos.com/models/picodet_s_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_s_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_s_320_coco_lcnet.yml) |\n+| PicoDet-S | 416*416 | 32.5 | 47.6 | 1.18 | 1.65 | 6.6ms | 15.20ms | [model](https://paddledet.bj.bcebos.com/models/picodet_s_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_s_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_s_416_coco_lcnet.yml) |\n+| PicoDet-M | 320*320 | 34.4 | 50.0 | 3.46 | 2.57 | 8.2ms | 17.68ms | [model](https://paddledet.bj.bcebos.com/models/picodet_m_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_m_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_m_320_coco_lcnet.yml) |\n+| PicoDet-M | 416*416 | 37.5 | 53.4 | 3.46 | 4.34 | 12.7ms | 28.39ms | [model](https://paddledet.bj.bcebos.com/models/picodet_m_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_m_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_m_416_coco_lcnet.yml) |\n+| PicoDet-L | 320*320 | 36.1 | 52.0 | 5.80 | 4.20 | 11.5ms | 25.21ms | [model](https://paddledet.bj.bcebos.com/models/picodet_l_320_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_l_320_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_l_320_coco_lcnet.yml) |\n+| PicoDet-L | 416*416 | 39.4 | 55.7 | 5.80 | 7.10 | 20.7ms | 42.23ms | [model](https://paddledet.bj.bcebos.com/models/picodet_l_416_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_l_416_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_l_416_coco_lcnet.yml) |\n+| PicoDet-L | 640*640 | 42.6 | 59.2 | 5.80 | 16.81 | 62.5ms | 108.1ms | [model](https://paddledet.bj.bcebos.com/models/picodet_l_640_coco_lcnet.pdparams) | [log](https://paddledet.bj.bcebos.com/logs/train_picodet_l_640_coco_lcnet.log) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/picodet/picodet_l_640_coco_lcnet.yml) |\n<details open>\n<summary><b>Table Notes:</b></summary>\n-- <a name=\"latency\">Latency:</a> All our models test on `Intel-Xeon-Gold-6148` CPU with MKLDNN by 10 threads and `Qualcomm Snapdragon 865(4xA77+4xA55)` with 4 threads by arm8 and with FP16. In the above table, test CPU latency on Paddle-Inference and testing Mobile latency with `Lite`->[Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite).\n+- <a name=\"latency\">Latency:</a> All our models test on `Intel core i7 10750H` CPU with MKLDNN by 12 threads and `Qualcomm Snapdragon 865(4xA77+4xA55)` with 4 threads by arm8 and with FP16. In the above table, test CPU latency on Paddle-Inference and testing Mobile latency with `Lite`->[Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite).\n- PicoDet is trained on COCO train2017 dataset and evaluated on COCO val2017. And PicoDet used 4 GPUs for training and all checkpoints are trained with default settings and hyperparameters.\n- Benchmark test: When testing the speed benchmark, the post-processing is not included in the exported model, you need to set `-o export.benchmark=True` or manually modify [runtime.yml](https://github.com/PaddlePaddle/PaddleDetection/blob/develop/configs/runtime.yml#L12).\n@@ -138,7 +138,7 @@ Detail also can refer to [Quick start guide](https://github.com/PaddlePaddle/Pad\ncd PaddleDetection\npython tools/export_model.py -c configs/picodet/picodet_s_320_coco_lcnet.yml \\\n-o weights=https://paddledet.bj.bcebos.com/models/picodet_s_320_coco_lcnet.pdparams \\\n- --output_dir=inference_model\n+ --output_dir=output_inference\n```\n- If no post processing is required, please specify: `-o export.benchmark=True` (if -o has already appeared, delete -o here) or manually modify corresponding fields in [runtime.yml](https://github.com/PaddlePaddle/PaddleDetection/blob/develop/configs/runtime.yml).\n@@ -160,9 +160,9 @@ pip install paddlelite\n```shell\n# FP32\n-paddle_lite_opt --model_dir=inference_model/picodet_s_320_coco_lcnet --valid_targets=arm --optimize_out=picodet_s_320_coco_fp32\n+paddle_lite_opt --model_dir=output_inference/picodet_s_320_coco_lcnet --valid_targets=arm --optimize_out=picodet_s_320_coco_fp32\n# FP16\n-paddle_lite_opt --model_dir=inference_model/picodet_s_320_coco_lcnet --valid_targets=arm --optimize_out=picodet_s_320_coco_fp16 --enable_fp16=true\n+paddle_lite_opt --model_dir=output_inference/picodet_s_320_coco_lcnet --valid_targets=arm --optimize_out=picodet_s_320_coco_fp16 --enable_fp16=true\n```\n</details>\n@@ -202,19 +202,17 @@ paddle2onnx --model_dir output_inference/picodet_s_320_coco_lcnet/ \\\n- Deploy models\n-| Model | Input size | ONNX | Paddle Lite(fp32) | Paddle Lite(fp16) |\n+| Model | Input size | ONNX(w/o postprocess) | Paddle Lite(fp32) | Paddle Lite(fp16) |\n| :-------- | :--------: | :---------------------: | :----------------: | :----------------: |\n-| PicoDet-S | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_320_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320_fp16.tar) |\n-| PicoDet-S | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_fp16.tar) |\n-| PicoDet-M | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_320_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320_fp16.tar) |\n-| PicoDet-M | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416_fp16.tar) |\n-| PicoDet-L | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_320_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320_fp16.tar) |\n-| PicoDet-L | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n+| PicoDet-XS | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n+| PicoDet-XS | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n+| PicoDet-S | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320_fp16.tar) |\n+| PicoDet-S | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_fp16.tar) |\n+| PicoDet-M | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320_fp16.tar) |\n+| PicoDet-M | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416_fp16.tar) |\n+| PicoDet-L | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320_fp16.tar) |\n+| PicoDet-L | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n| PicoDet-L | 640*640 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_640_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640_fp16.tar) |\n-| PicoDet-Shufflenetv2 1x | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_shufflenetv2_1x_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_shufflenetv2_1x.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_shufflenetv2_1x_fp16.tar) |\n-| PicoDet-MobileNetv3-large 1x | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_mobilenetv3_large_1x_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_mobilenetv3_large_1x.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_mobilenetv3_large_1x_fp16.tar) |\n-| PicoDet-LCNet 1.5x | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_lcnet_1_5x_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_lcnet_1_5x.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_lcnet_1_5x_fp16.tar) |\n-\n### Deploy\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_openvino/python/openvino_benchmark.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 cv2\n+import numpy as np\n+import time\n+import argparse\n+from openvino.runtime import Core\n+\n+\n+def image_preprocess_mobilenetv3(img_path, re_shape):\n+ img = cv2.imread(img_path)\n+ img = cv2.resize(\n+ img, (re_shape, re_shape), interpolation=cv2.INTER_LANCZOS4)\n+ img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n+ img = np.transpose(img, [2, 0, 1]) / 255\n+ img = np.expand_dims(img, 0)\n+ img_mean = np.array([0.485, 0.456, 0.406]).reshape((3, 1, 1))\n+ img_std = np.array([0.229, 0.224, 0.225]).reshape((3, 1, 1))\n+ img -= img_mean\n+ img /= img_std\n+ return img.astype(np.float32)\n+\n+\n+def benchmark(img_file, onnx_file, re_shape):\n+\n+ ie = Core()\n+ net = ie.read_model(onnx_file)\n+\n+ test_image = image_preprocess_mobilenetv3(img_file, re_shape)\n+\n+ compiled_model = ie.compile_model(net, 'CPU')\n+\n+ # benchmark\n+ loop_num = 100\n+ warm_up = 8\n+ timeall = 0\n+ time_min = float(\"inf\")\n+ time_max = float('-inf')\n+\n+ for i in range(loop_num + warm_up):\n+ time0 = time.time()\n+ #perform the inference step\n+\n+ output = compiled_model.infer_new_request({0: test_image})\n+ time1 = time.time()\n+ timed = time1 - time0\n+\n+ if i >= warm_up:\n+ timeall = timeall + timed\n+ time_min = min(time_min, timed)\n+ time_max = max(time_max, timed)\n+\n+ time_avg = timeall / loop_num\n+\n+ print(\n+ f'inference_time(ms): min={round(time_min*1000, 2)}, max = {round(time_max*1000, 1)}, avg = {round(time_avg*1000, 1)}'\n+ )\n+\n+\n+if __name__ == '__main__':\n+\n+ onnx_path = \"out_onnx\"\n+ onnx_file = onnx_path + \"/picodet_s_320_coco.onnx\"\n+\n+ parser = argparse.ArgumentParser()\n+ parser.add_argument(\n+ '--img_path',\n+ type=str,\n+ default='demo/000000570688.jpg',\n+ help=\"image path\")\n+ parser.add_argument(\n+ '--onnx_path',\n+ type=str,\n+ default='out_onnxsim/picodet_xs_320_coco_lcnet.onnx',\n+ help=\"onnx filepath\")\n+ parser.add_argument('--in_shape', type=int, default=320, help=\"input_size\")\n+\n+ args = parser.parse_args()\n+ benchmark(args.img_path, args.onnx_path, args.in_shape)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[picodet]openvino benchmark (#5578)
|
499,298 |
05.04.2022 21:08:51
| -28,800 |
cf1b5ce991cd435fdcf8a84d117e119d5d2eb919
|
[MOT] support xpu npu MOT infer
|
[
{
"change_type": "MODIFY",
"old_path": "tools/eval_mot.py",
"new_path": "tools/eval_mot.py",
"diff": "@@ -28,11 +28,11 @@ import warnings\nwarnings.filterwarnings('ignore')\nimport paddle\n-from paddle.distributed import ParallelEnv\n+\nfrom ppdet.core.workspace import load_config, merge_config\n-from ppdet.engine import Tracker\n-from ppdet.utils.check import check_gpu, check_version, check_config\n+from ppdet.utils.check import check_gpu, check_npu, check_xpu, check_version, check_config\nfrom ppdet.utils.cli import ArgsParser\n+from ppdet.engine import Tracker\ndef parse_args():\n@@ -104,9 +104,9 @@ def main():\ncfg = load_config(FLAGS.config)\nmerge_config(FLAGS.opt)\n- check_config(cfg)\n- check_gpu(cfg.use_gpu)\n- check_version()\n+ # disable npu in config by default\n+ if 'use_npu' not in cfg:\n+ cfg.use_npu = False\n# disable xpu in config by default\nif 'use_xpu' not in cfg:\n@@ -114,11 +114,22 @@ def main():\nif cfg.use_gpu:\nplace = paddle.set_device('gpu')\n+ elif cfg.use_npu:\n+ place = paddle.set_device('npu')\nelif cfg.use_xpu:\nplace = paddle.set_device('xpu')\nelse:\nplace = paddle.set_device('cpu')\n+ if 'norm_type' in cfg and cfg['norm_type'] == 'sync_bn' and not cfg.use_gpu:\n+ cfg['norm_type'] = 'bn'\n+\n+ check_config(cfg)\n+ check_gpu(cfg.use_gpu)\n+ check_npu(cfg.use_npu)\n+ check_xpu(cfg.use_xpu)\n+ check_version()\n+\nrun(FLAGS, cfg)\n"
},
{
"change_type": "MODIFY",
"old_path": "tools/infer_mot.py",
"new_path": "tools/infer_mot.py",
"diff": "@@ -28,10 +28,9 @@ import warnings\nwarnings.filterwarnings('ignore')\nimport paddle\n-from paddle.distributed import ParallelEnv\nfrom ppdet.core.workspace import load_config, merge_config\nfrom ppdet.engine import Tracker\n-from ppdet.utils.check import check_gpu, check_version, check_config\n+from ppdet.utils.check import check_gpu, check_npu, check_xpu, check_version, check_config\nfrom ppdet.utils.cli import ArgsParser\n@@ -117,12 +116,32 @@ def main():\ncfg = load_config(FLAGS.config)\nmerge_config(FLAGS.opt)\n+ # disable npu in config by default\n+ if 'use_npu' not in cfg:\n+ cfg.use_npu = False\n+\n+ # disable xpu in config by default\n+ if 'use_xpu' not in cfg:\n+ cfg.use_xpu = False\n+\n+ if cfg.use_gpu:\n+ place = paddle.set_device('gpu')\n+ elif cfg.use_npu:\n+ place = paddle.set_device('npu')\n+ elif cfg.use_xpu:\n+ place = paddle.set_device('xpu')\n+ else:\n+ place = paddle.set_device('cpu')\n+\n+ if 'norm_type' in cfg and cfg['norm_type'] == 'sync_bn' and not cfg.use_gpu:\n+ cfg['norm_type'] = 'bn'\n+\ncheck_config(cfg)\ncheck_gpu(cfg.use_gpu)\n+ check_npu(cfg.use_npu)\n+ check_xpu(cfg.use_xpu)\ncheck_version()\n- place = 'gpu:{}'.format(ParallelEnv().dev_id) if cfg.use_gpu else 'cpu'\n- place = paddle.set_device(place)\nrun(FLAGS, cfg)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[MOT] support xpu npu MOT infer (#5584)
|
499,304 |
06.04.2022 11:28:02
| -28,800 |
2fe432558ef1b497c5ae3f73bddb1ed4e22df3ae
|
update picodet inference model
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/legacy_model/README.md",
"new_path": "configs/picodet/legacy_model/README.md",
"diff": "</details>\n+- Deploy models\n+\n+| Model | Input size | ONNX | Paddle Lite(fp32) | Paddle Lite(fp16) |\n+| :-------- | :--------: | :---------------------: | :----------------: | :----------------: |\n+| PicoDet-S | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_320_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320_fp16.tar) |\n+| PicoDet-S | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_fp16.tar) |\n+| PicoDet-M | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_320_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320_fp16.tar) |\n+| PicoDet-M | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416_fp16.tar) |\n+| PicoDet-L | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_320_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320_fp16.tar) |\n+| PicoDet-L | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n+| PicoDet-L | 640*640 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_640_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640_fp16.tar) |\n+| PicoDet-Shufflenetv2 1x | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_shufflenetv2_1x_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_shufflenetv2_1x.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_shufflenetv2_1x_fp16.tar) |\n+| PicoDet-MobileNetv3-large 1x | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_mobilenetv3_large_1x_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_mobilenetv3_large_1x.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_mobilenetv3_large_1x_fp16.tar) |\n+| PicoDet-LCNet 1.5x | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_lcnet_1_5x_416_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_lcnet_1_5x.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_lcnet_1_5x_fp16.tar) |\n+\n+\n+\n## Cite PP-PicoDet\n```\n@misc{yu2021pppicodet,\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update picodet inference model (#5582)
|
499,304 |
06.04.2022 13:14:47
| -28,800 |
e23ab8d8e31d7c6d261d19d99c0a74bbed29d94a
|
fix picodet openvino demo
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_openvino/python/openvino_benchmark.py",
"new_path": "deploy/third_engine/demo_openvino/python/openvino_benchmark.py",
"diff": "@@ -19,7 +19,7 @@ import argparse\nfrom openvino.runtime import Core\n-def image_preprocess_mobilenetv3(img_path, re_shape):\n+def image_preprocess(img_path, re_shape):\nimg = cv2.imread(img_path)\nimg = cv2.resize(\nimg, (re_shape, re_shape), interpolation=cv2.INTER_LANCZOS4)\n@@ -38,7 +38,7 @@ def benchmark(img_file, onnx_file, re_shape):\nie = Core()\nnet = ie.read_model(onnx_file)\n- test_image = image_preprocess_mobilenetv3(img_file, re_shape)\n+ test_image = image_preprocess(img_file, re_shape)\ncompiled_model = ie.compile_model(net, 'CPU')\n@@ -64,9 +64,9 @@ def benchmark(img_file, onnx_file, re_shape):\ntime_avg = timeall / loop_num\n- print(\n- f'inference_time(ms): min={round(time_min*1000, 2)}, max = {round(time_max*1000, 1)}, avg = {round(time_avg*1000, 1)}'\n- )\n+ print('inference_time(ms): min={}, max={}, avg={}'.format(\n+ round(time_min * 1000, 2),\n+ round(time_max * 1000, 1), round(time_avg * 1000, 1)))\nif __name__ == '__main__':\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix picodet openvino demo (#5595)
|
499,333 |
06.04.2022 14:31:55
| -28,800 |
ae54352f57f397df6f9da5c2c87ee54aa486a9be
|
fix rcnn pred
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/architectures/faster_rcnn.py",
"new_path": "ppdet/modeling/architectures/faster_rcnn.py",
"diff": "@@ -87,8 +87,8 @@ class FasterRCNN(BaseArch):\nim_shape, scale_factor)\n# rescale the prediction back to origin image\n- bbox_pred = self.bbox_post_process.get_pred(bbox, bbox_num,\n- im_shape, scale_factor)\n+ bboxes, bbox_pred, bbox_num = self.bbox_post_process.get_pred(\n+ bbox, bbox_num, im_shape, scale_factor)\nreturn bbox_pred, bbox_num\ndef get_loss(self, ):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix rcnn pred (#5598)
|
499,298 |
06.04.2022 18:52:02
| -28,800 |
057ef8bd937913afacd0e27f3582d77bdcb47c80
|
add tqdm eval and infer
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/callbacks.py",
"new_path": "ppdet/engine/callbacks.py",
"diff": "@@ -141,10 +141,6 @@ class LogPrinter(Callback):\ndtime=str(data_time),\nips=ips)\nlogger.info(fmt)\n- if mode == 'eval':\n- step_id = status['step_id']\n- if step_id % 100 == 0:\n- logger.info(\"Eval iter: {}\".format(step_id))\ndef on_epoch_end(self, status):\nif dist.get_world_size() < 2 or dist.get_rank() == 0:\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/tracker.py",
"new_path": "ppdet/engine/tracker.py",
"diff": "@@ -17,12 +17,11 @@ from __future__ import division\nfrom __future__ import print_function\nimport os\n-import cv2\nimport glob\nimport re\nimport paddle\nimport numpy as np\n-import os.path as osp\n+from tqdm import tqdm\nfrom collections import defaultdict\nfrom ppdet.core.workspace import create\n@@ -31,8 +30,7 @@ from ppdet.modeling.mot.utils import Detection, get_crops, scale_coords, clip_bo\nfrom ppdet.modeling.mot.utils import MOTTimer, load_det_results, write_mot_results, save_vis_results\nfrom ppdet.modeling.mot.tracker import JDETracker, DeepSORTTracker\n-from ppdet.metrics import Metric, MOTMetric, KITTIMOTMetric\n-from ppdet.metrics import MCMOTMetric\n+from ppdet.metrics import Metric, MOTMetric, KITTIMOTMetric, MCMOTMetric\nimport ppdet.utils.stats as stats\nfrom .callbacks import Callback, ComposeCallback\n@@ -142,11 +140,8 @@ class Tracker(object):\nself.model.eval()\nresults = defaultdict(list) # support single class and multi classes\n- for step_id, data in enumerate(dataloader):\n+ for step_id, data in enumerate(tqdm(dataloader)):\nself.status['step_id'] = step_id\n- if frame_id % 40 == 0:\n- logger.info('Processing frame {} ({:.2f} fps)'.format(\n- frame_id, 1. / max(1e-5, timer.average_time)))\n# forward\ntimer.tic()\npred_dets, pred_embs = self.model(data)\n@@ -210,12 +205,8 @@ class Tracker(object):\ndet_file))\ntracker = self.model.tracker\n- for step_id, data in enumerate(dataloader):\n+ for step_id, data in enumerate(tqdm(dataloader)):\nself.status['step_id'] = step_id\n- if frame_id % 40 == 0:\n- logger.info('Processing frame {} ({:.2f} fps)'.format(\n- frame_id, 1. / max(1e-5, timer.average_time)))\n-\nori_image = data['ori_image'] # [bs, H, W, 3]\nori_image_shape = data['ori_image'].shape[1:3]\n# ori_image_shape: [H, W]\n@@ -366,8 +357,8 @@ class Tracker(object):\nonline_scores[cls_id].append(tscore)\n# save results\nresults[cls_id].append(\n- (frame_id + 1, online_tlwhs[cls_id], online_scores[cls_id],\n- online_ids[cls_id]))\n+ (frame_id + 1, online_tlwhs[cls_id],\n+ online_scores[cls_id], online_ids[cls_id]))\ntimer.toc()\nsave_vis_results(data, frame_id, online_ids, online_tlwhs,\nonline_scores, timer.average_time, show_image,\n@@ -417,7 +408,7 @@ class Tracker(object):\nsave_dir = os.path.join(output_dir, 'mot_outputs',\nseq) if save_images or save_videos else None\n- logger.info('start seq: {}'.format(seq))\n+ logger.info('Evaluate seq: {}'.format(seq))\nself.dataset.set_images(self.get_infer_images(infer_dir))\ndataloader = create('EvalMOTReader')(self.dataset, 0)\n@@ -458,7 +449,6 @@ class Tracker(object):\nos.system(cmd_str)\nlogger.info('Save video in {}.'.format(output_video_path))\n- logger.info('Evaluate seq: {}'.format(seq))\n# update metrics\nfor metric in self._metrics:\nmetric.update(data_root, seq, data_type, result_root,\n@@ -582,6 +572,7 @@ class Tracker(object):\nwrite_mot_results(result_filename, results, data_type,\nself.cfg.num_classes)\n+\ndef get_trick_hyperparams(video_name, ori_buffer, ori_thresh):\nif video_name[:3] != 'MOT':\n# only used for MOTChallenge (MOT17, MOT20) Test-set\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -20,6 +20,7 @@ import os\nimport sys\nimport copy\nimport time\n+from tqdm import tqdm\nimport numpy as np\nimport typing\n@@ -500,7 +501,7 @@ class Trainer(object):\nflops_loader = create('{}Reader'.format(self.mode.capitalize()))(\nself.dataset, self.cfg.worker_num, self._eval_batch_sampler)\nself._flops(flops_loader)\n- for step_id, data in enumerate(loader):\n+ for step_id, data in enumerate(tqdm(loader)):\nself.status['step_id'] = step_id\nself._compose_callback.on_step_begin(self.status)\n# forward\n@@ -553,7 +554,7 @@ class Trainer(object):\nflops_loader = create('TestReader')(self.dataset, 0)\nself._flops(flops_loader)\nresults = []\n- for step_id, data in enumerate(loader):\n+ for step_id, data in enumerate(tqdm(loader)):\nself.status['step_id'] = step_id\n# forward\nouts = self.model(data)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add tqdm eval and infer (#5587)
|
499,319 |
07.04.2022 09:42:00
| -28,800 |
bbece39537a8180bd4d6aa2dbceedd3118b35ffa
|
[pico] openvino demo
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_openvino/python/coco_label.txt",
"diff": "+person\n+bicycle\n+car\n+motorbike\n+aeroplane\n+bus\n+train\n+truck\n+boat\n+traffic light\n+fire hydrant\n+stop sign\n+parking meter\n+bench\n+bird\n+cat\n+dog\n+horse\n+sheep\n+cow\n+elephant\n+bear\n+zebra\n+giraffe\n+backpack\n+umbrella\n+handbag\n+tie\n+suitcase\n+frisbee\n+skis\n+snowboard\n+sports ball\n+kite\n+baseball bat\n+baseball glove\n+skateboard\n+surfboard\n+tennis racket\n+bottle\n+wine glass\n+cup\n+fork\n+knife\n+spoon\n+bowl\n+banana\n+apple\n+sandwich\n+orange\n+broccoli\n+carrot\n+hot dog\n+pizza\n+donut\n+cake\n+chair\n+sofa\n+pottedplant\n+bed\n+diningtable\n+toilet\n+tvmonitor\n+laptop\n+mouse\n+remote\n+keyboard\n+cell phone\n+microwave\n+oven\n+toaster\n+sink\n+refrigerator\n+book\n+clock\n+vase\n+scissors\n+teddy bear\n+hair drier\n+toothbrush\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_openvino/python/openvino_benchmark.py",
"new_path": "deploy/third_engine/demo_openvino/python/openvino_benchmark.py",
"diff": "@@ -16,6 +16,7 @@ import cv2\nimport numpy as np\nimport time\nimport argparse\n+from scipy.special import softmax\nfrom openvino.runtime import Core\n@@ -33,14 +34,275 @@ def image_preprocess(img_path, re_shape):\nreturn img.astype(np.float32)\n-def benchmark(img_file, onnx_file, re_shape):\n+def draw_box(img, results, class_label, scale_x, scale_y):\n- ie = Core()\n- net = ie.read_model(onnx_file)\n+ label_list = list(\n+ map(lambda x: x.strip(), open(class_label, 'r').readlines()))\n- test_image = image_preprocess(img_file, re_shape)\n+ for i in range(len(results)):\n+ print(label_list[int(results[i][0])], ':', results[i][1])\n+ bbox = results[i, 2:]\n+ label_id = int(results[i, 0])\n+ score = results[i, 1]\n+ if (score > 0.20):\n+ xmin, ymin, xmax, ymax = [\n+ int(bbox[0] * scale_x), int(bbox[1] * scale_y),\n+ int(bbox[2] * scale_x), int(bbox[3] * scale_y)\n+ ]\n+ cv2.rectangle(img, (xmin, ymin), (xmax, ymax), (0, 255, 0), 3)\n+ font = cv2.FONT_HERSHEY_SIMPLEX\n+ label_text = label_list[label_id]\n+ cv2.rectangle(img, (xmin, ymin), (xmax, ymin - 60), (0, 255, 0), -1)\n+ cv2.putText(img, \"#\" + label_text, (xmin, ymin - 10), font, 1,\n+ (255, 255, 255), 2, cv2.LINE_AA)\n+ cv2.putText(img,\n+ str(round(score, 3)), (xmin, ymin - 40), font, 0.8,\n+ (255, 255, 255), 2, cv2.LINE_AA)\n+ return img\n- compiled_model = ie.compile_model(net, 'CPU')\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 PicoDetPostProcess(object):\n+ \"\"\"\n+ Args:\n+ input_shape (int): network input image size\n+ ori_shape (int): ori image shape of before padding\n+ scale_factor (float): scale factor of ori image\n+ enable_mkldnn (bool): whether to open MKLDNN\n+ \"\"\"\n+\n+ def __init__(self,\n+ input_shape,\n+ ori_shape,\n+ scale_factor,\n+ strides=[8, 16, 32, 64],\n+ score_threshold=0.4,\n+ nms_threshold=0.5,\n+ nms_top_k=1000,\n+ keep_top_k=100):\n+ self.ori_shape = ori_shape\n+ self.input_shape = input_shape\n+ self.scale_factor = scale_factor\n+ self.strides = strides\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 warp_boxes(self, boxes, ori_shape):\n+ \"\"\"Apply transform to boxes\n+ \"\"\"\n+ width, height = ori_shape[1], ori_shape[0]\n+ n = len(boxes)\n+ if n:\n+ # warp points\n+ xy = np.ones((n * 4, 3))\n+ xy[:, :2] = boxes[:, [0, 1, 2, 3, 0, 3, 2, 1]].reshape(\n+ n * 4, 2) # x1y1, x2y2, x1y2, x2y1\n+ # xy = xy @ M.T # transform\n+ xy = (xy[:, :2] / xy[:, 2:3]).reshape(n, 8) # rescale\n+ # create new boxes\n+ x = xy[:, [0, 2, 4, 6]]\n+ y = xy[:, [1, 3, 5, 7]]\n+ xy = np.concatenate(\n+ (x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T\n+ # clip boxes\n+ xy[:, [0, 2]] = xy[:, [0, 2]].clip(0, width)\n+ xy[:, [1, 3]] = xy[:, [1, 3]].clip(0, height)\n+ return xy.astype(np.float32)\n+ else:\n+ return boxes\n+\n+ def __call__(self, scores, raw_boxes):\n+ batch_size = raw_boxes[0].shape[0]\n+ reg_max = int(raw_boxes[0].shape[-1] / 4 - 1)\n+ out_boxes_num = []\n+ out_boxes_list = []\n+ for batch_id in range(batch_size):\n+ # generate centers\n+ decode_boxes = []\n+ select_scores = []\n+ for stride, box_distribute, score in zip(self.strides, raw_boxes,\n+ scores):\n+ box_distribute = box_distribute[batch_id]\n+ score = score[batch_id]\n+ # centers\n+ fm_h = self.input_shape[0] / stride\n+ fm_w = self.input_shape[1] / stride\n+ h_range = np.arange(fm_h)\n+ w_range = np.arange(fm_w)\n+ ww, hh = np.meshgrid(w_range, h_range)\n+ ct_row = (hh.flatten() + 0.5) * stride\n+ ct_col = (ww.flatten() + 0.5) * stride\n+ center = np.stack((ct_col, ct_row, ct_col, ct_row), axis=1)\n+\n+ # box distribution to distance\n+ reg_range = np.arange(reg_max + 1)\n+ box_distance = box_distribute.reshape((-1, reg_max + 1))\n+ box_distance = softmax(box_distance, axis=1)\n+ box_distance = box_distance * np.expand_dims(reg_range, axis=0)\n+ box_distance = np.sum(box_distance, axis=1).reshape((-1, 4))\n+ box_distance = box_distance * stride\n+\n+ # top K candidate\n+ topk_idx = np.argsort(score.max(axis=1))[::-1]\n+ topk_idx = topk_idx[:self.nms_top_k]\n+ center = center[topk_idx]\n+ score = score[topk_idx]\n+ box_distance = box_distance[topk_idx]\n+\n+ # decode box\n+ decode_box = center + [-1, -1, 1, 1] * box_distance\n+\n+ select_scores.append(score)\n+ decode_boxes.append(decode_box)\n+\n+ # nms\n+ bboxes = np.concatenate(decode_boxes, axis=0)\n+ confidences = np.concatenate(select_scores, axis=0)\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.keep_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+ out_boxes_num.append(0)\n+\n+ else:\n+ picked_box_probs = np.concatenate(picked_box_probs)\n+\n+ # resize output boxes\n+ picked_box_probs[:, :4] = self.warp_boxes(\n+ picked_box_probs[:, :4], self.ori_shape[batch_id])\n+ im_scale = np.concatenate([\n+ self.scale_factor[batch_id][::-1],\n+ self.scale_factor[batch_id][::-1]\n+ ])\n+ picked_box_probs[:, :4] /= im_scale\n+ # clas score box\n+ out_boxes_list.append(\n+ np.concatenate(\n+ [\n+ np.expand_dims(\n+ np.array(picked_labels),\n+ axis=-1), np.expand_dims(\n+ picked_box_probs[:, 4], axis=-1),\n+ picked_box_probs[:, :4]\n+ ],\n+ axis=1))\n+ out_boxes_num.append(len(picked_labels))\n+\n+ out_boxes_list = np.concatenate(out_boxes_list, axis=0)\n+ out_boxes_num = np.asarray(out_boxes_num).astype(np.int32)\n+ return out_boxes_list, out_boxes_num\n+\n+\n+def detect(img_file, compiled_model, re_shape, class_label):\n+ output = compiled_model.infer_new_request({0: test_image})\n+ result_ie = list(output.values()) #[0]\n+\n+ test_im_shape = np.array([[re_shape, re_shape]]).astype('float32')\n+ test_scale_factor = np.array([[1, 1]]).astype('float32')\n+\n+ np_score_list = []\n+ np_boxes_list = []\n+\n+ num_outs = int(len(result_ie) / 2)\n+ for out_idx in range(num_outs):\n+ np_score_list.append(result_ie[out_idx])\n+ np_boxes_list.append(result_ie[out_idx + num_outs])\n+\n+ postprocess = PicoDetPostProcess(test_image.shape[2:], test_im_shape,\n+ test_scale_factor)\n+\n+ np_boxes, np_boxes_num = postprocess(np_score_list, np_boxes_list)\n+\n+ image = cv2.imread(img_file, 1)\n+ scale_x = image.shape[1] / test_image.shape[3]\n+ scale_y = image.shape[0] / test_image.shape[2]\n+ res_image = draw_box(image, np_boxes, class_label, scale_x, scale_y)\n+\n+ cv2.imwrite('res.jpg', res_image)\n+ cv2.imshow(\"res\", res_image)\n+ cv2.waitKey()\n+\n+\n+def benchmark(test_image, compiled_model):\n# benchmark\nloop_num = 100\n@@ -71,21 +333,33 @@ def benchmark(img_file, onnx_file, re_shape):\nif __name__ == '__main__':\n- onnx_path = \"out_onnx\"\n- onnx_file = onnx_path + \"/picodet_s_320_coco.onnx\"\n-\nparser = argparse.ArgumentParser()\n+ parser.add_argument(\n+ '--benchmark', type=int, default=1, help=\"0:detect; 1:benchmark\")\nparser.add_argument(\n'--img_path',\ntype=str,\n- default='demo/000000570688.jpg',\n+ default='demo/000000014439.jpg',\nhelp=\"image path\")\nparser.add_argument(\n'--onnx_path',\ntype=str,\n- default='out_onnxsim/picodet_xs_320_coco_lcnet.onnx',\n+ default='out_onnxsim/picodet_s_320_processed.onnx',\nhelp=\"onnx filepath\")\nparser.add_argument('--in_shape', type=int, default=320, help=\"input_size\")\n-\n+ parser.add_argument(\n+ '--class_label',\n+ type=str,\n+ default='coco_label.txt',\n+ help=\"class label file\")\nargs = parser.parse_args()\n- benchmark(args.img_path, args.onnx_path, args.in_shape)\n+\n+ ie = Core()\n+ net = ie.read_model(args.onnx_path)\n+ test_image = image_preprocess(args.img_path, args.in_shape)\n+ compiled_model = ie.compile_model(net, 'CPU')\n+\n+ if args.benchmark == 0:\n+ detect(args.img_path, compiled_model, args.in_shape, args.class_label)\n+ if args.benchmark == 1:\n+ benchmark(test_image, compiled_model)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[pico] openvino demo (#5605)
|
499,333 |
07.04.2022 14:52:00
| -28,800 |
2bce73cc6c27422a2fc4fcac7cce5fe6fe21034a
|
fix centernet deploy, test=develop
|
[
{
"change_type": "MODIFY",
"old_path": "configs/centernet/_base_/centernet_reader.yml",
"new_path": "configs/centernet/_base_/centernet_reader.yml",
"diff": "@@ -30,6 +30,6 @@ TestReader:\nsample_transforms:\n- Decode: {}\n- WarpAffine: {keep_res: True, input_h: 512, input_w: 512}\n- - NormalizeImage: {mean: [0.40789655, 0.44719303, 0.47026116], std: [0.2886383 , 0.27408165, 0.27809834]}\n+ - NormalizeImage: {mean: [0.40789655, 0.44719303, 0.47026116], std: [0.2886383 , 0.27408165, 0.27809834], is_scale: True}\n- Permute: {}\nbatch_size: 1\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/include/preprocess_op.h",
"new_path": "deploy/cpp/include/preprocess_op.h",
"diff": "@@ -74,7 +74,7 @@ class NormalizeImage : public PreprocessOp {\n// CHW or HWC\nstd::vector<float> mean_;\nstd::vector<float> scale_;\n- bool is_scale_;\n+ bool is_scale_ = true;\n};\nclass Permute : public PreprocessOp {\n@@ -143,6 +143,24 @@ class TopDownEvalAffine : public PreprocessOp {\nstd::vector<int> trainsize_;\n};\n+class WarpAffine : public PreprocessOp {\n+ public:\n+ virtual void Init(const YAML::Node& item) {\n+ input_h_ = item[\"input_h\"].as<int>();\n+ input_w_ = item[\"input_w\"].as<int>();\n+ keep_res_ = item[\"keep_res\"].as<bool>();\n+ }\n+\n+ virtual void Run(cv::Mat* im, ImageBlob* data);\n+\n+ private:\n+ int input_h_;\n+ int input_w_;\n+ int interp_ = 1;\n+ bool keep_res_ = true;\n+ int pad_ = 31;\n+};\n+\nvoid CropImg(cv::Mat& img,\ncv::Mat& crop_img,\nstd::vector<int>& area,\n@@ -183,6 +201,8 @@ class Preprocessor {\nreturn std::make_shared<PadStride>();\n} else if (name == \"TopDownEvalAffine\") {\nreturn std::make_shared<TopDownEvalAffine>();\n+ } else if (name == \"WarpAffine\") {\n+ return std::make_shared<WarpAffine>();\n}\nstd::cerr << \"can not find function of OP: \" << name\n<< \" and return: nullptr\" << std::endl;\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/src/preprocess_op.cc",
"new_path": "deploy/cpp/src/preprocess_op.cc",
"diff": "@@ -177,11 +177,64 @@ void TopDownEvalAffine::Run(cv::Mat* im, ImageBlob* data) {\n};\n}\n+void GetAffineTrans(const cv::Point2f center,\n+ const cv::Point2f input_size,\n+ const cv::Point2f output_size,\n+ cv::Mat* trans) {\n+ cv::Point2f srcTri[3];\n+ cv::Point2f dstTri[3];\n+ float src_w = input_size.x;\n+ float dst_w = output_size.x;\n+ float dst_h = output_size.y;\n+\n+ cv::Point2f src_dir(0, -0.5 * src_w);\n+ cv::Point2f dst_dir(0, -0.5 * dst_w);\n+\n+ srcTri[0] = center;\n+ srcTri[1] = center + src_dir;\n+ cv::Point2f src_d = srcTri[0] - srcTri[1];\n+ srcTri[2] = srcTri[1] + cv::Point2f(-src_d.y, src_d.x);\n+\n+ dstTri[0] = cv::Point2f(dst_w * 0.5, dst_h * 0.5);\n+ dstTri[1] = cv::Point2f(dst_w * 0.5, dst_h * 0.5) + dst_dir;\n+ cv::Point2f dst_d = dstTri[0] - dstTri[1];\n+ dstTri[2] = dstTri[1] + cv::Point2f(-dst_d.y, dst_d.x);\n+\n+ *trans = cv::getAffineTransform(srcTri, dstTri);\n+}\n+\n+void WarpAffine::Run(cv::Mat* im, ImageBlob* data) {\n+ cv::cvtColor(*im, *im, cv::COLOR_RGB2BGR);\n+ cv::Mat trans(2, 3, CV_32FC1);\n+ cv::Point2f center;\n+ cv::Point2f input_size;\n+ int h = im->rows;\n+ int w = im->cols;\n+ if (keep_res_) {\n+ input_h_ = (h | pad_) + 1;\n+ input_w_ = (w + pad_) + 1;\n+ input_size = cv::Point2f(input_w_, input_h_);\n+ center = cv::Point2f(w / 2, h / 2);\n+ } else {\n+ float s = std::max(h, w) * 1.0;\n+ input_size = cv::Point2f(s, s);\n+ center = cv::Point2f(w / 2., h / 2.);\n+ }\n+ cv::Point2f output_size(input_w_, input_h_);\n+\n+ GetAffineTrans(center, input_size, output_size, &trans);\n+ cv::warpAffine(*im, *im, trans, cv::Size(input_w_, input_h_));\n+ data->in_net_shape_ = {\n+ static_cast<float>(input_h_), static_cast<float>(input_w_),\n+ };\n+}\n+\n// Preprocessor op running order\nconst std::vector<std::string> Preprocessor::RUN_ORDER = {\"InitInfo\",\n\"TopDownEvalAffine\",\n\"Resize\",\n\"LetterBoxResize\",\n+ \"WarpAffine\",\n\"NormalizeImage\",\n\"PadStride\",\n\"Permute\"};\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/post_process.py",
"new_path": "ppdet/modeling/post_process.py",
"diff": "@@ -524,7 +524,7 @@ class CenterNetPostProcess(TTFBox):\nx2 = xs + wh[:, 0:1] / 2\ny2 = ys + wh[:, 1:2] / 2\n- n, c, feat_h, feat_w = hm.shape[:]\n+ n, c, feat_h, feat_w = paddle.shape(hm)\npadw = (feat_w * self.down_ratio - im_shape[0, 1]) / 2\npadh = (feat_h * self.down_ratio - im_shape[0, 0]) / 2\nx1 = x1 * self.down_ratio\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix centernet deploy, test=develop (#5616)
|
499,304 |
07.04.2022 18:48:00
| -28,800 |
573d7314da78d6e279c4c28921bdc36da3dbed91
|
add picodet quant model
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "configs/slim/quant/picodet_s_416_lcnet_quant.yml",
"diff": "+pretrain_weights: https://paddledet.bj.bcebos.com/models/picodet_s_416_coco_lcnet.pdparams\n+slim: QAT\n+\n+QAT:\n+ quant_config: {\n+ 'activation_preprocess_type': 'PACT',\n+ 'weight_quantize_type': 'channel_wise_abs_max', 'activation_quantize_type': 'moving_average_abs_max',\n+ 'weight_bits': 8, 'activation_bits': 8, 'dtype': 'int8', 'window_size': 10000, 'moving_rate': 0.9,\n+ 'quantizable_layer_type': ['Conv2D', 'Linear']}\n+ print_model: False\n+\n+TrainReader:\n+ batch_size: 48\n+\n+LearningRate:\n+ base_lr: 0.024\n+ schedulers:\n+ - !CosineDecay\n+ max_epochs: 300\n+ - !LinearWarmup\n+ start_factor: 0.1\n+ steps: 300\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add picodet quant model (#5617)
|
499,339 |
08.04.2022 17:40:40
| -28,800 |
dfba7a5d81d240aed03be5c04afcff421d516276
|
[PPYOLOE] fix doc, test=document_fix
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/README.md",
"new_path": "configs/ppyoloe/README.md",
"diff": "@@ -99,13 +99,13 @@ python tools/export_model.py -c configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml -o w\n```bash\n# inference single image\n-CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyolo_r50vd_dcn_1x_coco --image_file=demo/000000014439_640x640.jpg --device=gpu\n+CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyoloe_crn_l_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu\n# inference all images in the directory\n-CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyolo_r50vd_dcn_1x_coco --image_dir=demo/ --device=gpu\n+CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyoloe_crn_l_300e_coco --image_dir=demo/ --device=gpu\n# benchmark\n-CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyolo_r50vd_dcn_1x_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True\n+CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/ppyoloe_crn_l_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True\n```\nIf you want to export PP-YOLOE model to **ONNX format**, use following command refer to [PaddleDetection Model Export as ONNX Format Tutorial](../../deploy/EXPORT_ONNX_MODEL_en.md).\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[PPYOLOE] fix doc, test=document_fix (#5633)
|
499,339 |
08.04.2022 20:10:49
| -28,800 |
7fbea77a791a9528cb3e6d709c587e395a8de9b4
|
[TIPC] fix log path in shell
|
[
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_train_inference_python.sh",
"new_path": "test_tipc/test_train_inference_python.sh",
"diff": "@@ -92,7 +92,7 @@ benchmark_value=$(func_parser_value \"${lines[49]}\")\ninfer_key1=$(func_parser_key \"${lines[50]}\")\ninfer_value1=$(func_parser_value \"${lines[50]}\")\n-LOG_PATH=\"./test_tipc/output\"\n+LOG_PATH=\"./test_tipc/output/${model_name}\"\nmkdir -p ${LOG_PATH}\nstatus_log=\"${LOG_PATH}/results_python.log\"\n@@ -305,7 +305,7 @@ else\nstatus_check $? \"${export_cmd}\" \"${status_log}\"\n#run inference\n- save_export_model_dir=\"${save_export_value}/${model_name}\"\n+ save_export_model_dir=\"${save_log}/${model_name}\"\neval $env\nfunc_inference \"${python}\" \"${inference_py}\" \"${save_export_model_dir}\" \"${LOG_PATH}\" \"${train_infer_img_dir}\" \"${flag_quant}\"\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[TIPC] fix log path in shell (#5638)
|
499,319 |
09.04.2022 10:12:59
| -28,800 |
ce2a78dd89c8fde1cab1d2d91115a1cca4d5bad2
|
[pico] add onnxruntime demo
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/README_en.md",
"new_path": "configs/picodet/README_en.md",
"diff": "@@ -204,15 +204,16 @@ paddle2onnx --model_dir output_inference/picodet_s_320_coco_lcnet/ \\\n| Model | Input size | ONNX(w/o postprocess) | Paddle Lite(fp32) | Paddle Lite(fp16) |\n| :-------- | :--------: | :---------------------: | :----------------: | :----------------: |\n-| PicoDet-XS | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n-| PicoDet-XS | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n-| PicoDet-S | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320_fp16.tar) |\n-| PicoDet-S | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_fp16.tar) |\n-| PicoDet-M | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320_fp16.tar) |\n-| PicoDet-M | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416_fp16.tar) |\n-| PicoDet-L | 320*320 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320_fp16.tar) |\n-| PicoDet-L | 416*416 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n-| PicoDet-L | 640*640 | [model](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_640_coco.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640_fp16.tar) |\n+| PicoDet-XS | 320*320 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_320_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_xs_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_xs_320_fp16.tar) |\n+| PicoDet-XS | 416*416 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_416_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_xs_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_xs_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_xs_416_fp16.tar) |\n+| PicoDet-S | 320*320 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_320_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_320_fp16.tar) |\n+| PicoDet-S | 416*416 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_416_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_s_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_s_416_fp16.tar) |\n+| PicoDet-M | 320*320 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_320_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_320_fp16.tar) |\n+| PicoDet-M | 416*416 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_416_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_m_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_m_416_fp16.tar) |\n+| PicoDet-L | 320*320 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_320_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_320_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_320_fp16.tar) |\n+| PicoDet-L | 416*416 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_416_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_416_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_416_fp16.tar) |\n+| PicoDet-L | 640*640 | [( w/ postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_640_lcnet_postprocessed.onnx) | [( w/o postprocess)](https://paddledet.bj.bcebos.com/deploy/third_engine/picodet_l_640_coco_lcnet.onnx) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640.tar) | [model](https://paddledet.bj.bcebos.com/deploy/paddlelite/picodet_l_640_fp16.tar) |\n+\n### Deploy\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_onnxruntime/coco_label.txt",
"diff": "+person\n+bicycle\n+car\n+motorbike\n+aeroplane\n+bus\n+train\n+truck\n+boat\n+traffic light\n+fire hydrant\n+stop sign\n+parking meter\n+bench\n+bird\n+cat\n+dog\n+horse\n+sheep\n+cow\n+elephant\n+bear\n+zebra\n+giraffe\n+backpack\n+umbrella\n+handbag\n+tie\n+suitcase\n+frisbee\n+skis\n+snowboard\n+sports ball\n+kite\n+baseball bat\n+baseball glove\n+skateboard\n+surfboard\n+tennis racket\n+bottle\n+wine glass\n+cup\n+fork\n+knife\n+spoon\n+bowl\n+banana\n+apple\n+sandwich\n+orange\n+broccoli\n+carrot\n+hot dog\n+pizza\n+donut\n+cake\n+chair\n+sofa\n+pottedplant\n+bed\n+diningtable\n+toilet\n+tvmonitor\n+laptop\n+mouse\n+remote\n+keyboard\n+cell phone\n+microwave\n+oven\n+toaster\n+sink\n+refrigerator\n+book\n+clock\n+vase\n+scissors\n+teddy bear\n+hair drier\n+toothbrush\n"
},
{
"change_type": "ADD",
"old_path": "deploy/third_engine/demo_onnxruntime/imgs/bus.jpg",
"new_path": "deploy/third_engine/demo_onnxruntime/imgs/bus.jpg",
"diff": "Binary files /dev/null and b/deploy/third_engine/demo_onnxruntime/imgs/bus.jpg differ\n"
},
{
"change_type": "ADD",
"old_path": "deploy/third_engine/demo_onnxruntime/imgs/dog.jpg",
"new_path": "deploy/third_engine/demo_onnxruntime/imgs/dog.jpg",
"diff": "Binary files /dev/null and b/deploy/third_engine/demo_onnxruntime/imgs/dog.jpg differ\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_onnxruntime/infer_demo.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 cv2\n+import numpy as np\n+import argparse\n+import onnxruntime as ort\n+from pathlib import Path\n+from tqdm import tqdm\n+\n+\n+class PicoDet():\n+ def __init__(self,\n+ model_pb_path,\n+ label_path,\n+ prob_threshold=0.4,\n+ iou_threshold=0.3):\n+ self.classes = list(\n+ map(lambda x: x.strip(), open(label_path, 'r').readlines()))\n+ self.num_classes = len(self.classes)\n+ self.prob_threshold = prob_threshold\n+ self.iou_threshold = iou_threshold\n+ self.mean = np.array(\n+ [103.53, 116.28, 123.675], dtype=np.float32).reshape(1, 1, 3)\n+ self.std = np.array(\n+ [57.375, 57.12, 58.395], dtype=np.float32).reshape(1, 1, 3)\n+ so = ort.SessionOptions()\n+ so.log_severity_level = 3\n+ self.net = ort.InferenceSession(model_pb_path, so)\n+ self.input_shape = (self.net.get_inputs()[0].shape[2],\n+ self.net.get_inputs()[0].shape[3])\n+\n+ def _normalize(self, img):\n+ img = img.astype(np.float32)\n+ img = (img / 255.0 - self.mean / 255.0) / (self.std / 255.0)\n+ return img\n+\n+ def resize_image(self, srcimg, keep_ratio=False):\n+ top, left, newh, neww = 0, 0, self.input_shape[0], self.input_shape[1]\n+ origin_shape = srcimg.shape[:2]\n+ im_scale_y = newh / float(origin_shape[0])\n+ im_scale_x = neww / float(origin_shape[1])\n+ scale_factor = np.array([[im_scale_y, im_scale_x]]).astype('float32')\n+\n+ if keep_ratio and srcimg.shape[0] != srcimg.shape[1]:\n+ hw_scale = srcimg.shape[0] / srcimg.shape[1]\n+ if hw_scale > 1:\n+ newh, neww = self.input_shape[0], int(self.input_shape[1] /\n+ hw_scale)\n+ img = cv2.resize(\n+ srcimg, (neww, newh), interpolation=cv2.INTER_AREA)\n+ left = int((self.input_shape[1] - neww) * 0.5)\n+ img = cv2.copyMakeBorder(\n+ img,\n+ 0,\n+ 0,\n+ left,\n+ self.input_shape[1] - neww - left,\n+ cv2.BORDER_CONSTANT,\n+ value=0) # add border\n+ else:\n+ newh, neww = int(self.input_shape[0] *\n+ hw_scale), self.input_shape[1]\n+ img = cv2.resize(\n+ srcimg, (neww, newh), interpolation=cv2.INTER_AREA)\n+ top = int((self.input_shape[0] - newh) * 0.5)\n+ img = cv2.copyMakeBorder(\n+ img,\n+ top,\n+ self.input_shape[0] - newh - top,\n+ 0,\n+ 0,\n+ cv2.BORDER_CONSTANT,\n+ value=0)\n+ else:\n+ img = cv2.resize(\n+ srcimg, self.input_shape, interpolation=cv2.INTER_AREA)\n+\n+ return img, scale_factor\n+\n+ def get_color_map_list(self, num_classes):\n+ color_map = num_classes * [0, 0, 0]\n+ for i in range(0, num_classes):\n+ j = 0\n+ lab = i\n+ while lab:\n+ color_map[i * 3] |= (((lab >> 0) & 1) << (7 - j))\n+ color_map[i * 3 + 1] |= (((lab >> 1) & 1) << (7 - j))\n+ color_map[i * 3 + 2] |= (((lab >> 2) & 1) << (7 - j))\n+ j += 1\n+ lab >>= 3\n+ color_map = [color_map[i:i + 3] for i in range(0, len(color_map), 3)]\n+ return color_map\n+\n+ def detect(self, srcimg):\n+ img, scale_factor = self.resize_image(srcimg)\n+ img = self._normalize(img)\n+\n+ blob = np.expand_dims(np.transpose(img, (2, 0, 1)), axis=0)\n+\n+ outs = self.net.run(None, {\n+ self.net.get_inputs()[0].name: blob,\n+ self.net.get_inputs()[1].name: scale_factor\n+ })\n+\n+ outs = np.array(outs[0])\n+ expect_boxes = (outs[:, 1] > 0.5) & (outs[:, 0] > -1)\n+ np_boxes = outs[expect_boxes, :]\n+\n+ color_list = self.get_color_map_list(self.num_classes)\n+ clsid2color = {}\n+\n+ for i in range(np_boxes.shape[0]):\n+ classid, conf = int(np_boxes[i, 0]), np_boxes[i, 1]\n+ xmin, ymin, xmax, ymax = int(np_boxes[i, 2]), int(np_boxes[\n+ i, 3]), int(np_boxes[i, 4]), int(np_boxes[i, 5])\n+\n+ if classid not in clsid2color:\n+ clsid2color[classid] = color_list[classid]\n+ color = tuple(clsid2color[classid])\n+\n+ cv2.rectangle(\n+ srcimg, (xmin, ymin), (xmax, ymax), color, thickness=2)\n+ print(self.classes[classid] + ': ' + str(round(conf, 3)))\n+ cv2.putText(\n+ srcimg,\n+ self.classes[classid] + ':' + str(round(conf, 3)), (xmin,\n+ ymin - 10),\n+ cv2.FONT_HERSHEY_SIMPLEX,\n+ 0.8, (0, 255, 0),\n+ thickness=2)\n+\n+ return srcimg\n+\n+ def detect_folder(self, img_fold, result_path):\n+ img_fold = Path(img_fold)\n+ result_path = Path(result_path)\n+ result_path.mkdir(parents=True, exist_ok=True)\n+\n+ img_name_list = filter(\n+ lambda x: str(x).endswith(\".png\") or str(x).endswith(\".jpg\"),\n+ img_fold.iterdir(), )\n+ img_name_list = list(img_name_list)\n+ print(f\"find {len(img_name_list)} images\")\n+\n+ for img_path in tqdm(img_name_list):\n+ img = cv2.imread(str(img_path))\n+\n+ srcimg = net.detect(img)\n+ save_path = str(result_path / img_path.name.replace(\".png\", \".jpg\"))\n+ cv2.imwrite(save_path, srcimg)\n+\n+\n+if __name__ == '__main__':\n+ parser = argparse.ArgumentParser()\n+ parser.add_argument(\n+ '--modelpath',\n+ type=str,\n+ default='onnx_file/picodet_s_320_lcnet_postprocessed.onnx',\n+ help=\"onnx filepath\")\n+ parser.add_argument(\n+ '--classfile',\n+ type=str,\n+ default='coco_label.txt',\n+ help=\"classname filepath\")\n+ parser.add_argument(\n+ '--confThreshold', default=0.5, type=float, help='class confidence')\n+ parser.add_argument(\n+ '--nmsThreshold', default=0.6, type=float, help='nms iou thresh')\n+ parser.add_argument(\n+ \"--img_fold\", dest=\"img_fold\", type=str, default=\"./imgs\")\n+ parser.add_argument(\n+ \"--result_fold\", dest=\"result_fold\", type=str, default=\"./results\")\n+ args = parser.parse_args()\n+\n+ net = PicoDet(\n+ args.modelpath,\n+ args.classfile,\n+ prob_threshold=args.confThreshold,\n+ iou_threshold=args.nmsThreshold)\n+\n+ net.detect_folder(args.img_fold, args.result_fold)\n"
},
{
"change_type": "ADD",
"old_path": "docs/images/bus.jpg",
"new_path": "docs/images/bus.jpg",
"diff": "Binary files /dev/null and b/docs/images/bus.jpg differ\n"
},
{
"change_type": "ADD",
"old_path": "docs/images/dog.jpg",
"new_path": "docs/images/dog.jpg",
"diff": "Binary files /dev/null and b/docs/images/dog.jpg differ\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[pico] add onnxruntime demo (#5627)
|
499,395 |
11.04.2022 19:52:58
| -28,800 |
bc5d5245bf991be48b3e60e46301affdce33b21f
|
remove register_buffer
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/ppyoloe_head.py",
"new_path": "ppdet/modeling/heads/ppyoloe_head.py",
"diff": "@@ -132,8 +132,8 @@ class PPYOLOEHead(nn.Layer):\nif self.eval_size:\nanchor_points, stride_tensor = self._generate_anchors()\n- self.register_buffer('anchor_points', anchor_points)\n- self.register_buffer('stride_tensor', stride_tensor)\n+ self.anchor_points = anchor_points\n+ self.stride_tensor = stride_tensor\ndef forward_train(self, feats, targets):\nanchors, anchor_points, num_anchors_list, stride_tensor = \\\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
remove register_buffer (#5631)
|
499,314 |
12.04.2022 10:29:47
| -28,800 |
df4a27c6db2e237c47c63b888905d2a1f4803fba
|
change core.ops to _C_ops
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/ops.py",
"new_path": "ppdet/modeling/ops.py",
"diff": "@@ -17,6 +17,7 @@ import paddle.nn.functional as F\nimport paddle.nn as nn\nfrom paddle import ParamAttr\nfrom paddle.regularizer import L2Decay\n+from paddle import _C_ops\nfrom paddle.fluid.framework import Variable, in_dygraph_mode\nfrom paddle.fluid import core\n@@ -174,7 +175,7 @@ def roi_pool(input,\npooled_height, pooled_width = output_size\nif in_dygraph_mode():\nassert rois_num is not None, \"rois_num should not be None in dygraph mode.\"\n- pool_out, argmaxes = core.ops.roi_pool(\n+ pool_out, argmaxes = _C_ops.roi_pool(\ninput, rois, rois_num, \"pooled_height\", pooled_height,\n\"pooled_width\", pooled_width, \"spatial_scale\", spatial_scale)\nreturn pool_out, argmaxes\n@@ -281,7 +282,7 @@ def roi_align(input,\nif in_dygraph_mode():\nassert rois_num is not None, \"rois_num should not be None in dygraph mode.\"\n- align_out = core.ops.roi_align(\n+ align_out = _C_ops.roi_align(\ninput, rois, rois_num, \"pooled_height\", pooled_height,\n\"pooled_width\", pooled_width, \"spatial_scale\", spatial_scale,\n\"sampling_ratio\", sampling_ratio, \"aligned\", aligned)\n@@ -364,7 +365,7 @@ def iou_similarity(x, y, box_normalized=True, name=None):\n\"\"\"\nif in_dygraph_mode():\n- out = core.ops.iou_similarity(x, y, 'box_normalized', box_normalized)\n+ out = _C_ops.iou_similarity(x, y, 'box_normalized', box_normalized)\nreturn out\nelse:\nhelper = LayerHelper(\"iou_similarity\", **locals())\n@@ -460,7 +461,7 @@ def collect_fpn_proposals(multi_rois,\nif in_dygraph_mode():\nassert rois_num_per_level is not None, \"rois_num_per_level should not be None in dygraph mode.\"\nattrs = ('post_nms_topN', post_nms_top_n)\n- output_rois, rois_num = core.ops.collect_fpn_proposals(\n+ output_rois, rois_num = _C_ops.collect_fpn_proposals(\ninput_rois, input_scores, rois_num_per_level, *attrs)\nreturn output_rois, rois_num\n@@ -575,7 +576,7 @@ def distribute_fpn_proposals(fpn_rois,\nattrs = ('min_level', min_level, 'max_level', max_level, 'refer_level',\nrefer_level, 'refer_scale', refer_scale, 'pixel_offset',\npixel_offset)\n- multi_rois, restore_ind, rois_num_per_level = core.ops.distribute_fpn_proposals(\n+ multi_rois, restore_ind, rois_num_per_level = _C_ops.distribute_fpn_proposals(\nfpn_rois, rois_num, num_lvl, num_lvl, *attrs)\nreturn multi_rois, restore_ind, rois_num_per_level\n@@ -729,7 +730,7 @@ def yolo_box(\nattrs = ('anchors', anchors, 'class_num', class_num, 'conf_thresh',\nconf_thresh, 'downsample_ratio', downsample_ratio, 'clip_bbox',\nclip_bbox, 'scale_x_y', scale_x_y)\n- boxes, scores = core.ops.yolo_box(x, origin_shape, *attrs)\n+ boxes, scores = _C_ops.yolo_box(x, origin_shape, *attrs)\nreturn boxes, scores\nelse:\nboxes = helper.create_variable_for_type_inference(dtype=x.dtype)\n@@ -867,7 +868,7 @@ def prior_box(input,\n'min_max_aspect_ratios_order', min_max_aspect_ratios_order)\nif cur_max_sizes is not None:\nattrs += ('max_sizes', cur_max_sizes)\n- box, var = core.ops.prior_box(input, image, *attrs)\n+ box, var = _C_ops.prior_box(input, image, *attrs)\nreturn box, var\nelse:\nattrs = {\n@@ -1010,7 +1011,7 @@ def multiclass_nms(bboxes,\nscore_threshold, 'nms_top_k', nms_top_k, 'nms_threshold',\nnms_threshold, 'keep_top_k', keep_top_k, 'nms_eta', nms_eta,\n'normalized', normalized)\n- output, index, nms_rois_num = core.ops.multiclass_nms3(bboxes, scores,\n+ output, index, nms_rois_num = _C_ops.multiclass_nms3(bboxes, scores,\nrois_num, *attrs)\nif not return_index:\nindex = None\n@@ -1152,7 +1153,7 @@ def matrix_nms(bboxes,\nnms_top_k, 'gaussian_sigma', gaussian_sigma, 'use_gaussian',\nuse_gaussian, 'keep_top_k', keep_top_k, 'normalized',\nnormalized)\n- out, index, rois_num = core.ops.matrix_nms(bboxes, scores, *attrs)\n+ out, index, rois_num = _C_ops.matrix_nms(bboxes, scores, *attrs)\nif not return_index:\nindex = None\nif not return_rois_num:\n@@ -1273,7 +1274,7 @@ def bipartite_match(dist_matrix,\n['float32', 'float64'], 'bipartite_match')\nif in_dygraph_mode():\n- match_indices, match_distance = core.ops.bipartite_match(\n+ match_indices, match_distance = _C_ops.bipartite_match(\ndist_matrix, \"match_type\", match_type, \"dist_threshold\",\ndist_threshold)\nreturn match_indices, match_distance\n@@ -1410,12 +1411,12 @@ def box_coder(prior_box,\nif in_dygraph_mode():\nif isinstance(prior_box_var, Variable):\n- output_box = core.ops.box_coder(\n+ output_box = _C_ops.box_coder(\nprior_box, prior_box_var, target_box, \"code_type\", code_type,\n\"box_normalized\", box_normalized, \"axis\", axis)\nelif isinstance(prior_box_var, list):\n- output_box = core.ops.box_coder(\n+ output_box = _C_ops.box_coder(\nprior_box, None, target_box, \"code_type\", code_type,\n\"box_normalized\", box_normalized, \"axis\", axis, \"variance\",\nprior_box_var)\n@@ -1538,7 +1539,7 @@ def generate_proposals(scores,\nattrs = ('pre_nms_topN', pre_nms_top_n, 'post_nms_topN', post_nms_top_n,\n'nms_thresh', nms_thresh, 'min_size', min_size, 'eta', eta,\n'pixel_offset', pixel_offset)\n- rpn_rois, rpn_roi_probs, rpn_rois_num = core.ops.generate_proposals_v2(\n+ rpn_rois, rpn_roi_probs, rpn_rois_num = _C_ops.generate_proposals_v2(\nscores, bbox_deltas, im_shape, anchors, variances, *attrs)\nif not return_rois_num:\nrpn_rois_num = None\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
change core.ops to _C_ops (#5663)
|
499,304 |
12.04.2022 11:58:14
| -28,800 |
c612935d8d7431f3a730cf5e213159f6b20938d1
|
Simplify picodet postprocess
|
[
{
"change_type": "MODIFY",
"old_path": "configs/picodet/_base_/picodet_320_reader.yml",
"new_path": "configs/picodet/_base_/picodet_320_reader.yml",
"diff": "worker_num: 6\n+eval_height: &eval_height 320\n+eval_width: &eval_width 320\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -18,7 +22,7 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [320, 320], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_transforms:\n@@ -29,13 +33,10 @@ EvalReader:\nTestReader:\ninputs_def:\n- image_shape: [1, 3, 320, 320]\n+ image_shape: [1, 3, *eval_height, *eval_width]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [320, 320], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\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\n- shuffle: false\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/picodet/_base_/picodet_416_reader.yml",
"new_path": "configs/picodet/_base_/picodet_416_reader.yml",
"diff": "worker_num: 6\n+eval_height: &eval_height 416\n+eval_width: &eval_width 416\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -18,7 +22,7 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [416, 416], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_transforms:\n@@ -29,13 +33,10 @@ EvalReader:\nTestReader:\ninputs_def:\n- image_shape: [1, 3, 416, 416]\n+ image_shape: [1, 3, *eval_height, *eval_width]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [416, 416], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\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\n- shuffle: false\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/picodet/_base_/picodet_640_reader.yml",
"new_path": "configs/picodet/_base_/picodet_640_reader.yml",
"diff": "worker_num: 6\n+eval_height: &eval_height 640\n+eval_width: &eval_width 640\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -18,7 +22,7 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [640, 640], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_transforms:\n@@ -29,13 +33,10 @@ EvalReader:\nTestReader:\ninputs_def:\n- image_shape: [1, 3, 640, 640]\n+ image_shape: [1, 3, *eval_height, *eval_width]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [640, 640], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\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\n- shuffle: false\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/picodet/legacy_model/_base_/picodet_320_reader.yml",
"new_path": "configs/picodet/legacy_model/_base_/picodet_320_reader.yml",
"diff": "worker_num: 6\n+eval_height: &eval_height 320\n+eval_width: &eval_width 320\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -18,7 +22,7 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [320, 320], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_transforms:\n@@ -29,13 +33,10 @@ EvalReader:\nTestReader:\ninputs_def:\n- image_shape: [1, 3, 320, 320]\n+ image_shape: [1, 3, *eval_height, *eval_width]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [320, 320], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\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\n- shuffle: false\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/picodet/legacy_model/_base_/picodet_416_reader.yml",
"new_path": "configs/picodet/legacy_model/_base_/picodet_416_reader.yml",
"diff": "worker_num: 6\n+eval_height: &eval_height 416\n+eval_width: &eval_width 416\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -18,7 +22,7 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [416, 416], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_transforms:\n@@ -29,13 +33,10 @@ EvalReader:\nTestReader:\ninputs_def:\n- image_shape: [1, 3, 416, 416]\n+ image_shape: [1, 3, *eval_height, *eval_width]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [416, 416], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\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\n- shuffle: false\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/picodet/legacy_model/_base_/picodet_640_reader.yml",
"new_path": "configs/picodet/legacy_model/_base_/picodet_640_reader.yml",
"diff": "worker_num: 6\n+eval_height: &eval_height 640\n+eval_width: &eval_width 640\n+eval_size: &eval_size [*eval_height, *eval_width]\n+\nTrainReader:\nsample_transforms:\n- Decode: {}\n@@ -18,7 +22,7 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [640, 640], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\n- NormalizeImage: {is_scale: true, mean: [0.485,0.456,0.406], std: [0.229, 0.224,0.225]}\n- Permute: {}\nbatch_transforms:\n@@ -29,13 +33,10 @@ EvalReader:\nTestReader:\ninputs_def:\n- image_shape: [1, 3, 640, 640]\n+ image_shape: [1, 3, *eval_height, *eval_width]\nsample_transforms:\n- Decode: {}\n- - Resize: {interp: 2, target_size: [640, 640], keep_ratio: False}\n+ - Resize: {interp: 2, target_size: *eval_size, keep_ratio: False}\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\n- shuffle: false\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/architectures/picodet.py",
"new_path": "ppdet/modeling/architectures/picodet.py",
"diff": "@@ -67,10 +67,9 @@ class PicoDet(BaseArch):\nif self.training or not self.export_post_process:\nreturn head_outs, None\nelse:\n- im_shape = self.inputs['im_shape']\nscale_factor = self.inputs['scale_factor']\nbboxes, bbox_num = self.head.post_process(\n- head_outs, im_shape, scale_factor, export_nms=self.export_nms)\n+ head_outs, scale_factor, export_nms=self.export_nms)\nreturn bboxes, bbox_num\ndef get_loss(self, ):\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/gfl_head.py",
"new_path": "ppdet/modeling/heads/gfl_head.py",
"diff": "@@ -79,7 +79,9 @@ class Integral(nn.Layer):\noffsets from the box center in four directions, shape (N, 4).\n\"\"\"\nx = F.softmax(x.reshape([-1, self.reg_max + 1]), axis=1)\n- x = F.linear(x, self.project).reshape([-1, 4])\n+ x = F.linear(x, self.project)\n+ if self.training:\n+ x = x.reshape([-1, 4])\nreturn x\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/pico_head.py",
"new_path": "ppdet/modeling/heads/pico_head.py",
"diff": "@@ -194,7 +194,7 @@ class PicoHead(OTAVFLHead):\n'conv_feat', 'dgqp_module', 'loss_class', 'loss_dfl', 'loss_bbox',\n'assigner', 'nms'\n]\n- __shared__ = ['num_classes']\n+ __shared__ = ['num_classes', 'eval_size']\ndef __init__(self,\nconv_feat='PicoFeat',\n@@ -210,7 +210,8 @@ class PicoHead(OTAVFLHead):\nfeat_in_chan=96,\nnms=None,\nnms_pre=1000,\n- cell_offset=0):\n+ cell_offset=0,\n+ eval_size=None):\nsuper(PicoHead, self).__init__(\nconv_feat=conv_feat,\ndgqp_module=dgqp_module,\n@@ -239,6 +240,7 @@ class PicoHead(OTAVFLHead):\nself.nms = nms\nself.nms_pre = nms_pre\nself.cell_offset = cell_offset\n+ self.eval_size = eval_size\nself.use_sigmoid = self.loss_vfl.use_sigmoid\nif self.use_sigmoid:\n@@ -282,12 +284,50 @@ class PicoHead(OTAVFLHead):\nbias_attr=ParamAttr(initializer=Constant(value=0))))\nself.head_reg_list.append(head_reg)\n+ # initialize the anchor points\n+ if self.eval_size:\n+ self.anchor_points, self.stride_tensor = self._generate_anchors()\n+\ndef forward(self, fpn_feats, export_post_process=True):\nassert len(fpn_feats) == len(\nself.fpn_stride\n), \"The size of fpn_feats is not equal to size of fpn_stride\"\n- cls_logits_list = []\n- bboxes_reg_list = []\n+\n+ if self.training:\n+ return self.forward_train(fpn_feats)\n+ else:\n+ return self.forward_eval(\n+ fpn_feats, export_post_process=export_post_process)\n+\n+ def forward_train(self, fpn_feats):\n+ cls_logits_list, bboxes_reg_list = [], []\n+ for i, fpn_feat in enumerate(fpn_feats):\n+ conv_cls_feat, conv_reg_feat = self.conv_feat(fpn_feat, i)\n+ if self.conv_feat.share_cls_reg:\n+ cls_logits = self.head_cls_list[i](conv_cls_feat)\n+ cls_score, bbox_pred = paddle.split(\n+ cls_logits,\n+ [self.cls_out_channels, 4 * (self.reg_max + 1)],\n+ axis=1)\n+ else:\n+ cls_score = self.head_cls_list[i](conv_cls_feat)\n+ bbox_pred = self.head_reg_list[i](conv_reg_feat)\n+\n+ if self.dgqp_module:\n+ quality_score = self.dgqp_module(bbox_pred)\n+ cls_score = F.sigmoid(cls_score) * quality_score\n+\n+ cls_logits_list.append(cls_score)\n+ bboxes_reg_list.append(bbox_pred)\n+\n+ return (cls_logits_list, bboxes_reg_list)\n+\n+ def forward_eval(self, fpn_feats, export_post_process=True):\n+ if self.eval_size:\n+ anchor_points, stride_tensor = self.anchor_points, self.stride_tensor\n+ else:\n+ anchor_points, stride_tensor = self._generate_anchors(fpn_feats)\n+ cls_logits_list, bboxes_reg_list = [], []\nfor i, fpn_feat in enumerate(fpn_feats):\nconv_cls_feat, conv_reg_feat = self.conv_feat(fpn_feat, i)\nif self.conv_feat.share_cls_reg:\n@@ -307,50 +347,68 @@ class PicoHead(OTAVFLHead):\nif not export_post_process:\n# Now only supports batch size = 1 in deploy\n# TODO(ygh): support batch size > 1\n- cls_score = F.sigmoid(cls_score).reshape(\n+ cls_score_out = F.sigmoid(cls_score).reshape(\n[1, self.cls_out_channels, -1]).transpose([0, 2, 1])\nbbox_pred = bbox_pred.reshape([1, (self.reg_max + 1) * 4,\n-1]).transpose([0, 2, 1])\n- elif not self.training:\n- cls_score = F.sigmoid(cls_score.transpose([0, 2, 3, 1]))\n+ else:\n+ b, _, h, w = fpn_feat.shape\n+ l = h * w\n+ cls_score_out = F.sigmoid(\n+ cls_score.reshape([b, self.cls_out_channels, l]))\nbbox_pred = bbox_pred.transpose([0, 2, 3, 1])\n- stride = self.fpn_stride[i]\n- b, cell_h, cell_w, _ = paddle.shape(cls_score)\n- y, x = self.get_single_level_center_point(\n- [cell_h, cell_w], stride, cell_offset=self.cell_offset)\n- center_points = paddle.stack([x, y], axis=-1)\n- cls_score = cls_score.reshape([b, -1, self.cls_out_channels])\n- bbox_pred = self.distribution_project(bbox_pred) * stride\n- bbox_pred = bbox_pred.reshape([b, cell_h * cell_w, 4])\n-\n- # NOTE: If keep_ratio=False and image shape value that\n- # multiples of 32, distance2bbox not set max_shapes parameter\n- # to speed up model prediction. If need to set max_shapes,\n- # please use inputs['im_shape'].\n- bbox_pred = batch_distance2bbox(\n- center_points, bbox_pred, max_shapes=None)\n+ bbox_pred = self.distribution_project(bbox_pred)\n+ bbox_pred = bbox_pred.reshape([b, l, 4])\n- cls_logits_list.append(cls_score)\n+ cls_logits_list.append(cls_score_out)\nbboxes_reg_list.append(bbox_pred)\n+ if export_post_process:\n+ cls_logits_list = paddle.concat(cls_logits_list, axis=-1)\n+ bboxes_reg_list = paddle.concat(bboxes_reg_list, axis=1)\n+ bboxes_reg_list = batch_distance2bbox(anchor_points,\n+ bboxes_reg_list)\n+ bboxes_reg_list *= stride_tensor\n+\nreturn (cls_logits_list, bboxes_reg_list)\n- def post_process(self,\n- gfl_head_outs,\n- im_shape,\n- scale_factor,\n- export_nms=True):\n- cls_scores, bboxes_reg = gfl_head_outs\n- bboxes = paddle.concat(bboxes_reg, axis=1)\n- mlvl_scores = paddle.concat(cls_scores, axis=1)\n- mlvl_scores = mlvl_scores.transpose([0, 2, 1])\n+ def _generate_anchors(self, feats=None):\n+ # just use in eval time\n+ anchor_points = []\n+ stride_tensor = []\n+ for i, stride in enumerate(self.fpn_stride):\n+ if feats is not None:\n+ _, _, h, w = feats[i].shape\n+ else:\n+ h = math.ceil(self.eval_size[0] / stride)\n+ w = math.ceil(self.eval_size[1] / stride)\n+ shift_x = paddle.arange(end=w) + self.cell_offset\n+ shift_y = paddle.arange(end=h) + self.cell_offset\n+ shift_y, shift_x = paddle.meshgrid(shift_y, shift_x)\n+ anchor_point = paddle.cast(\n+ paddle.stack(\n+ [shift_x, shift_y], axis=-1), dtype='float32')\n+ anchor_points.append(anchor_point.reshape([-1, 2]))\n+ stride_tensor.append(\n+ paddle.full(\n+ [h * w, 1], stride, dtype='float32'))\n+ anchor_points = paddle.concat(anchor_points)\n+ stride_tensor = paddle.concat(stride_tensor)\n+ return anchor_points, stride_tensor\n+\n+ def post_process(self, head_outs, scale_factor, export_nms=True):\n+ pred_scores, pred_bboxes = head_outs\nif not export_nms:\n- return bboxes, mlvl_scores\n+ return pred_bboxes, pred_scores\nelse:\n# rescale: [h_scale, w_scale] -> [w_scale, h_scale, w_scale, h_scale]\n- im_scale = scale_factor.flip([1]).tile([1, 2]).unsqueeze(1)\n- bboxes /= im_scale\n- bbox_pred, bbox_num, _ = self.nms(bboxes, mlvl_scores)\n+ scale_y, scale_x = paddle.split(scale_factor, 2, axis=-1)\n+ scale_factor = paddle.concat(\n+ [scale_x, scale_y, scale_x, scale_y],\n+ axis=-1).reshape([-1, 1, 4])\n+ # scale bbox to origin image size.\n+ pred_bboxes /= scale_factor\n+ bbox_pred, bbox_num, _ = self.nms(pred_bboxes, pred_scores)\nreturn bbox_pred, bbox_num\n@@ -374,10 +432,9 @@ class PicoHeadV2(GFLHead):\n'conv_feat', 'dgqp_module', 'loss_class', 'loss_dfl', 'loss_bbox',\n'static_assigner', 'assigner', 'nms'\n]\n- __shared__ = ['num_classes']\n+ __shared__ = ['num_classes', 'eval_size']\n- def __init__(\n- self,\n+ def __init__(self,\nconv_feat='PicoFeatV2',\ndgqp_module=None,\nnum_classes=80,\n@@ -396,7 +453,8 @@ class PicoHeadV2(GFLHead):\nnms_pre=1000,\ncell_offset=0,\nact='hard_swish',\n- grid_cell_scale=5.0, ):\n+ grid_cell_scale=5.0,\n+ eval_size=None):\nsuper(PicoHeadV2, self).__init__(\nconv_feat=conv_feat,\ndgqp_module=dgqp_module,\n@@ -432,6 +490,7 @@ class PicoHeadV2(GFLHead):\nself.grid_cell_scale = grid_cell_scale\nself.use_align_head = use_align_head\nself.cls_out_channels = self.num_classes\n+ self.eval_size = eval_size\nbias_init_value = -math.log((1 - self.prior_prob) / self.prior_prob)\n# Clear the super class initialization\n@@ -478,11 +537,22 @@ class PicoHeadV2(GFLHead):\nact=self.act,\nuse_act_in_out=False))\n+ # initialize the anchor points\n+ if self.eval_size:\n+ self.anchor_points, self.stride_tensor = self._generate_anchors()\n+\ndef forward(self, fpn_feats, export_post_process=True):\nassert len(fpn_feats) == len(\nself.fpn_stride\n), \"The size of fpn_feats is not equal to size of fpn_stride\"\n+ if self.training:\n+ return self.forward_train(fpn_feats)\n+ else:\n+ return self.forward_eval(\n+ fpn_feats, export_post_process=export_post_process)\n+\n+ def forward_train(self, fpn_feats):\ncls_score_list, reg_list, box_list = [], [], []\nfor i, (fpn_feat, stride) in enumerate(zip(fpn_feats, self.fpn_stride)):\nb, _, h, w = get_static_shape(fpn_feat)\n@@ -498,15 +568,6 @@ class PicoHeadV2(GFLHead):\nelse:\ncls_score = F.sigmoid(cls_logit)\n- if not export_post_process and not self.training:\n- # Now only supports batch size = 1 in deploy\n- cls_score_list.append(\n- cls_score.reshape([1, self.cls_out_channels, -1]).transpose(\n- [0, 2, 1]))\n- box_list.append(\n- reg_pred.reshape([1, (self.reg_max + 1) * 4, -1]).transpose(\n- [0, 2, 1]))\n- else:\ncls_score_out = cls_score.transpose([0, 2, 3, 1])\nbbox_pred = reg_pred.transpose([0, 2, 3, 1])\nb, cell_h, cell_w, _ = paddle.shape(cls_score_out)\n@@ -519,23 +580,60 @@ class PicoHeadV2(GFLHead):\nbbox_pred = bbox_pred.reshape([b, cell_h * cell_w, 4])\nbbox_pred = batch_distance2bbox(\ncenter_points, bbox_pred, max_shapes=None)\n- if not self.training:\n- cls_score_list.append(cls_score_out)\n- box_list.append(bbox_pred)\n- else:\n- cls_score_list.append(\n- cls_score.flatten(2).transpose([0, 2, 1]))\n+ cls_score_list.append(cls_score.flatten(2).transpose([0, 2, 1]))\nreg_list.append(reg_pred.flatten(2).transpose([0, 2, 1]))\nbox_list.append(bbox_pred / stride)\n- if not self.training:\n- return cls_score_list, box_list\n- else:\ncls_score_list = paddle.concat(cls_score_list, axis=1)\nbox_list = paddle.concat(box_list, axis=1)\nreg_list = paddle.concat(reg_list, axis=1)\nreturn cls_score_list, reg_list, box_list, fpn_feats\n+ def forward_eval(self, fpn_feats, export_post_process=True):\n+ if self.eval_size:\n+ anchor_points, stride_tensor = self.anchor_points, self.stride_tensor\n+ else:\n+ anchor_points, stride_tensor = self._generate_anchors(fpn_feats)\n+ cls_score_list, box_list = [], []\n+ for i, (fpn_feat, stride) in enumerate(zip(fpn_feats, self.fpn_stride)):\n+ b, _, h, w = fpn_feat.shape\n+ # task decomposition\n+ conv_cls_feat, se_feat = self.conv_feat(fpn_feat, i)\n+ cls_logit = self.head_cls_list[i](se_feat)\n+ reg_pred = self.head_reg_list[i](se_feat)\n+\n+ # cls prediction and alignment\n+ if self.use_align_head:\n+ cls_prob = F.sigmoid(self.cls_align[i](conv_cls_feat))\n+ cls_score = (F.sigmoid(cls_logit) * cls_prob + eps).sqrt()\n+ else:\n+ cls_score = F.sigmoid(cls_logit)\n+\n+ if not export_post_process:\n+ # Now only supports batch size = 1 in deploy\n+ cls_score_list.append(\n+ cls_score.reshape([1, self.cls_out_channels, -1]).transpose(\n+ [0, 2, 1]))\n+ box_list.append(\n+ reg_pred.reshape([1, (self.reg_max + 1) * 4, -1]).transpose(\n+ [0, 2, 1]))\n+ else:\n+ l = h * w\n+ cls_score_out = cls_score.reshape([b, self.cls_out_channels, l])\n+ bbox_pred = reg_pred.transpose([0, 2, 3, 1])\n+ bbox_pred = self.distribution_project(bbox_pred)\n+ bbox_pred = bbox_pred.reshape([b, l, 4])\n+ cls_score_list.append(cls_score_out)\n+ box_list.append(bbox_pred)\n+\n+ if export_post_process:\n+ cls_score_list = paddle.concat(cls_score_list, axis=-1)\n+ box_list = paddle.concat(box_list, axis=1)\n+ box_list = batch_distance2bbox(anchor_points, box_list)\n+ box_list *= stride_tensor\n+\n+ return cls_score_list, box_list\n+\ndef get_loss(self, head_outs, gt_meta):\npred_scores, pred_regs, pred_bboxes, fpn_feats = head_outs\ngt_labels = gt_meta['gt_class']\n@@ -644,20 +742,41 @@ class PicoHeadV2(GFLHead):\nreturn loss_states\n- def post_process(self,\n- gfl_head_outs,\n- im_shape,\n- scale_factor,\n- export_nms=True):\n- cls_scores, bboxes_reg = gfl_head_outs\n- bboxes = paddle.concat(bboxes_reg, axis=1)\n- mlvl_scores = paddle.concat(cls_scores, axis=1)\n- mlvl_scores = mlvl_scores.transpose([0, 2, 1])\n+ def _generate_anchors(self, feats=None):\n+ # just use in eval time\n+ anchor_points = []\n+ stride_tensor = []\n+ for i, stride in enumerate(self.fpn_stride):\n+ if feats is not None:\n+ _, _, h, w = feats[i].shape\n+ else:\n+ h = math.ceil(self.eval_size[0] / stride)\n+ w = math.ceil(self.eval_size[1] / stride)\n+ shift_x = paddle.arange(end=w) + self.cell_offset\n+ shift_y = paddle.arange(end=h) + self.cell_offset\n+ shift_y, shift_x = paddle.meshgrid(shift_y, shift_x)\n+ anchor_point = paddle.cast(\n+ paddle.stack(\n+ [shift_x, shift_y], axis=-1), dtype='float32')\n+ anchor_points.append(anchor_point.reshape([-1, 2]))\n+ stride_tensor.append(\n+ paddle.full(\n+ [h * w, 1], stride, dtype='float32'))\n+ anchor_points = paddle.concat(anchor_points)\n+ stride_tensor = paddle.concat(stride_tensor)\n+ return anchor_points, stride_tensor\n+\n+ def post_process(self, head_outs, scale_factor, export_nms=True):\n+ pred_scores, pred_bboxes = head_outs\nif not export_nms:\n- return bboxes, mlvl_scores\n+ return pred_bboxes, pred_scores\nelse:\n# rescale: [h_scale, w_scale] -> [w_scale, h_scale, w_scale, h_scale]\n- im_scale = scale_factor.flip([1]).tile([1, 2]).unsqueeze(1)\n- bboxes /= im_scale\n- bbox_pred, bbox_num, _ = self.nms(bboxes, mlvl_scores)\n+ scale_y, scale_x = paddle.split(scale_factor, 2, axis=-1)\n+ scale_factor = paddle.concat(\n+ [scale_x, scale_y, scale_x, scale_y],\n+ axis=-1).reshape([-1, 1, 4])\n+ # scale bbox to origin image size.\n+ pred_bboxes /= scale_factor\n+ bbox_pred, bbox_num, _ = self.nms(pred_bboxes, pred_scores)\nreturn bbox_pred, bbox_num\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
Simplify picodet postprocess (#5650)
|
499,339 |
12.04.2022 16:05:18
| -28,800 |
0c4166af9e5a0f5adf1069d503cd0d666cad6174
|
fix assigner dtype error
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/assigners/atss_assigner.py",
"new_path": "ppdet/modeling/assigners/atss_assigner.py",
"diff": "@@ -124,7 +124,8 @@ class ATSSAssigner(nn.Layer):\n# negative batch\nif num_max_boxes == 0:\n- assigned_labels = paddle.full([batch_size, num_anchors], bg_index)\n+ assigned_labels = paddle.full(\n+ [batch_size, num_anchors], bg_index, dtype=gt_labels.dtype)\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": "@@ -85,7 +85,8 @@ class TaskAlignedAssigner(nn.Layer):\n# negative batch\nif num_max_boxes == 0:\n- assigned_labels = paddle.full([batch_size, num_anchors], bg_index)\n+ assigned_labels = paddle.full(\n+ [batch_size, num_anchors], bg_index, dtype=gt_labels.dtype)\nassigned_bboxes = paddle.zeros([batch_size, num_anchors, 4])\nassigned_scores = paddle.zeros(\n[batch_size, num_anchors, num_classes])\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix assigner dtype error (#5665)
|
499,299 |
13.04.2022 15:46:26
| -28,800 |
8dfe9bc7d71923879720bfb6b3c201fbc99c8e93
|
fix visualize error in keypoint infer
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/python/keypoint_infer.py",
"new_path": "deploy/python/keypoint_infer.py",
"diff": "@@ -267,8 +267,10 @@ class KeyPointDetector(Detector):\nprint('detect frame: %d' % (index))\nindex += 1\nresults = self.predict_image([frame], visual=False)\n+ im_results = {}\n+ im_results['keypoint'] = [results['keypoint'], results['score']]\nim = visualize_pose(\n- frame, results, visual_thresh=self.threshold, returnimg=True)\n+ frame, im_results, visual_thresh=self.threshold, returnimg=True)\nwriter.write(im)\nif camera_id != -1:\ncv2.imshow('Mask Detection', im)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix visualize error in keypoint infer (#5676)
|
499,319 |
14.04.2022 12:10:09
| -28,800 |
cfc020341c3209534e542dbdf161d0099117e396
|
[pico] openvino infer
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_openvino/python/openvino_benchmark.py",
"new_path": "deploy/third_engine/demo_openvino/python/openvino_benchmark.py",
"diff": "@@ -339,7 +339,7 @@ if __name__ == '__main__':\nparser.add_argument(\n'--img_path',\ntype=str,\n- default='demo/000000014439.jpg',\n+ default='../../../../demo/000000014439.jpg',\nhelp=\"image path\")\nparser.add_argument(\n'--onnx_path',\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "deploy/third_engine/demo_openvino/python/openvino_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 cv2\n+import numpy as np\n+import argparse\n+from scipy.special import softmax\n+from openvino.runtime import Core\n+\n+\n+def image_preprocess(img_path, re_shape):\n+ img = cv2.imread(img_path)\n+ img = cv2.resize(\n+ img, (re_shape, re_shape), interpolation=cv2.INTER_LANCZOS4)\n+ img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n+ img = np.transpose(img, [2, 0, 1]) / 255\n+ img = np.expand_dims(img, 0)\n+ img_mean = np.array([0.485, 0.456, 0.406]).reshape((3, 1, 1))\n+ img_std = np.array([0.229, 0.224, 0.225]).reshape((3, 1, 1))\n+ img -= img_mean\n+ img /= img_std\n+ return img.astype(np.float32)\n+\n+\n+def get_color_map_list(num_classes):\n+ color_map = num_classes * [0, 0, 0]\n+ for i in range(0, num_classes):\n+ j = 0\n+ lab = i\n+ while lab:\n+ color_map[i * 3] |= (((lab >> 0) & 1) << (7 - j))\n+ color_map[i * 3 + 1] |= (((lab >> 1) & 1) << (7 - j))\n+ color_map[i * 3 + 2] |= (((lab >> 2) & 1) << (7 - j))\n+ j += 1\n+ lab >>= 3\n+ color_map = [color_map[i:i + 3] for i in range(0, len(color_map), 3)]\n+ return color_map\n+\n+\n+def draw_box(srcimg, results, class_label):\n+ label_list = list(\n+ map(lambda x: x.strip(), open(class_label, 'r').readlines()))\n+ for i in range(len(results)):\n+ color_list = get_color_map_list(len(label_list))\n+ clsid2color = {}\n+ classid, conf = int(results[i, 0]), results[i, 1]\n+ xmin, ymin, xmax, ymax = int(results[i, 2]), int(results[i, 3]), int(\n+ results[i, 4]), int(results[i, 5])\n+\n+ if classid not in clsid2color:\n+ clsid2color[classid] = color_list[classid]\n+ color = tuple(clsid2color[classid])\n+\n+ cv2.rectangle(srcimg, (xmin, ymin), (xmax, ymax), color, thickness=2)\n+ print(label_list[classid] + ': ' + str(round(conf, 3)))\n+ cv2.putText(\n+ srcimg,\n+ label_list[classid] + ':' + str(round(conf, 3)), (xmin, ymin - 10),\n+ cv2.FONT_HERSHEY_SIMPLEX,\n+ 0.8, (0, 255, 0),\n+ thickness=2)\n+ return srcimg\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 PicoDetNMS(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+ input_shape,\n+ scale_x,\n+ scale_y,\n+ strides=[8, 16, 32, 64],\n+ score_threshold=0.4,\n+ nms_threshold=0.5,\n+ nms_top_k=1000,\n+ keep_top_k=100):\n+ self.input_shape = input_shape\n+ self.scale_x = scale_x\n+ self.scale_y = scale_y\n+ self.strides = strides\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 __call__(self, decode_boxes, select_scores):\n+ batch_size = 1\n+ out_boxes_list = []\n+ for batch_id in range(batch_size):\n+ # nms\n+ bboxes = np.concatenate(decode_boxes, axis=0)\n+ confidences = np.concatenate(select_scores, axis=0)\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.keep_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+\n+ # resize output boxes\n+ picked_box_probs[:, 0] *= self.scale_x\n+ picked_box_probs[:, 2] *= self.scale_x\n+ picked_box_probs[:, 1] *= self.scale_y\n+ picked_box_probs[:, 3] *= self.scale_y\n+\n+ # clas score box\n+ out_boxes_list.append(\n+ np.concatenate(\n+ [\n+ np.expand_dims(\n+ np.array(picked_labels),\n+ axis=-1), np.expand_dims(\n+ picked_box_probs[:, 4], axis=-1),\n+ picked_box_probs[:, :4]\n+ ],\n+ axis=1))\n+\n+ out_boxes_list = np.concatenate(out_boxes_list, axis=0)\n+ return out_boxes_list\n+\n+\n+def detect(img_file, compiled_model, class_label):\n+ output = compiled_model.infer_new_request({0: test_image})\n+ result_ie = list(output.values())\n+\n+ decode_boxes = []\n+ select_scores = []\n+ num_outs = int(len(result_ie) / 2)\n+ for out_idx in range(num_outs):\n+ decode_boxes.append(result_ie[out_idx])\n+ select_scores.append(result_ie[out_idx + num_outs])\n+\n+ image = cv2.imread(img_file, 1)\n+ scale_x = image.shape[1] / test_image.shape[3]\n+ scale_y = image.shape[0] / test_image.shape[2]\n+\n+ nms = PicoDetNMS(test_image.shape[2:], scale_x, scale_y)\n+ np_boxes = nms(decode_boxes, select_scores)\n+\n+ res_image = draw_box(image, np_boxes, class_label)\n+\n+ cv2.imwrite('res.jpg', res_image)\n+ cv2.imshow(\"res\", res_image)\n+ cv2.waitKey()\n+\n+\n+if __name__ == '__main__':\n+\n+ parser = argparse.ArgumentParser()\n+ parser.add_argument(\n+ '--img_path',\n+ type=str,\n+ default='../../demo_onnxruntime/imgs/bus.jpg',\n+ help=\"image path\")\n+ parser.add_argument(\n+ '--onnx_path',\n+ type=str,\n+ default='out_onnxsim_infer/picodet_s_320_postproccesed_woNMS.onnx',\n+ help=\"onnx filepath\")\n+ parser.add_argument('--in_shape', type=int, default=320, help=\"input_size\")\n+ parser.add_argument(\n+ '--class_label',\n+ type=str,\n+ default='coco_label.txt',\n+ help=\"class label file\")\n+ args = parser.parse_args()\n+\n+ ie = Core()\n+ net = ie.read_model(args.onnx_path)\n+ test_image = image_preprocess(args.img_path, args.in_shape)\n+ compiled_model = ie.compile_model(net, 'CPU')\n+\n+ detect(args.img_path, compiled_model, args.class_label)\n"
},
{
"change_type": "ADD",
"old_path": "docs/images/res.jpg",
"new_path": "docs/images/res.jpg",
"diff": "Binary files /dev/null and b/docs/images/res.jpg differ\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[pico] openvino infer (#5687)
|
499,304 |
14.04.2022 12:23:42
| -28,800 |
12865d90fc22c4ea65921ea8a4607a1b454ed573
|
fix Mask-RCNN QAT bug
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/post_process.py",
"new_path": "ppdet/modeling/post_process.py",
"diff": "@@ -33,7 +33,7 @@ __all__ = [\n@register\n-class BBoxPostProcess(nn.Layer):\n+class BBoxPostProcess(object):\n__shared__ = ['num_classes', 'export_onnx']\n__inject__ = ['decode', 'nms']\n@@ -45,7 +45,7 @@ class BBoxPostProcess(nn.Layer):\nself.nms = nms\nself.export_onnx = export_onnx\n- def forward(self, head_out, rois, im_shape, scale_factor):\n+ def __call__(self, head_out, rois, im_shape, scale_factor):\n\"\"\"\nDecode the bbox and do NMS if needed.\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix Mask-RCNN QAT bug (#5689)
|
499,301 |
15.04.2022 13:44:00
| -28,800 |
376c25c0955799cbbe10628cf0fc85010dabb41a
|
update readme, add yoloe
|
[
{
"change_type": "MODIFY",
"old_path": "README_en.md",
"new_path": "README_en.md",
"diff": "@@ -243,7 +243,7 @@ The relationship between COCO mAP and FPS on Tesla V100 of representative models\n- `PP-YOLO` achieves mAP of 45.9% on COCO and 72.9FPS on Tesla V100. Both precision and speed surpass [YOLOv4](https://arxiv.org/abs/2004.10934)\n- `PP-YOLO v2` is optimized version of `PP-YOLO` which has mAP of 49.5% and 68.9FPS on Tesla V100\n-\n+- `PP-YOLOE` is optimized version of `PP-YOLO v2` which has mAP of 51.4% and 78.1FPS on Tesla V100\n- All these models can be get in [Model Zoo](#ModelZoo)\nThe relationship between COCO mAP and FPS on Qualcomm Snapdragon 865 of representative mobile side models.\n"
},
{
"change_type": "MODIFY",
"old_path": "docs/images/fps_map.png",
"new_path": "docs/images/fps_map.png",
"diff": "Binary files a/docs/images/fps_map.png and b/docs/images/fps_map.png differ\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update readme, add yoloe (#5703)
|
499,339 |
20.04.2022 18:59:39
| -28,800 |
c0cb9b02be7ff7bea26b17a17a5a359c4a22090a
|
[PPYOLOE] fix loss bug in negative batch
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/ppyoloe_head.py",
"new_path": "ppdet/modeling/heads/ppyoloe_head.py",
"diff": "@@ -293,7 +293,7 @@ class PPYOLOEHead(nn.Layer):\nelse:\nloss_l1 = paddle.zeros([1])\nloss_iou = paddle.zeros([1])\n- loss_dfl = paddle.zeros([1])\n+ loss_dfl = pred_dist.sum() * 0.\nreturn loss_l1, loss_iou, loss_dfl\ndef get_loss(self, head_outs, gt_meta):\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[PPYOLOE] fix loss bug in negative batch (#5774)
|
499,313 |
21.04.2022 13:57:03
| -28,800 |
5652dca372899acfbacf46d9d8427bd1ce215250
|
fix build wheel script, run unittest separatly
|
[
{
"change_type": "MODIFY",
"old_path": "scripts/build_wheel.sh",
"new_path": "scripts/build_wheel.sh",
"diff": "@@ -91,9 +91,8 @@ function unittest() {\nif [ $? != 0 ]; then\nexit 1\nfi\n- find \"../ppdet\" -name 'tests' -type d -print0 | \\\n- xargs -0 -I{} -n1 bash -c \\\n- 'python -m unittest discover -v -s {}'\n+ find \"../ppdet\" -wholename '*tests/test_*' -type f -print0 | \\\n+ xargs -0 -I{} -n1 -t bash -c 'python -u -s {}'\n# clean TEST_DIR\ncd ..\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix build wheel script, run unittest separatly (#5776)
|
499,298 |
22.04.2022 10:40:35
| -28,800 |
96959b21641544f2f334946e707cdf78738c03af
|
refine RetinaNet codes
|
[
{
"change_type": "MODIFY",
"old_path": "configs/retinanet/README.md",
"new_path": "configs/retinanet/README.md",
"diff": "-# Focal Loss for Dense Object Detection\n-\n-## Introduction\n-\n-We reproduce RetinaNet proposed in paper Focal Loss for Dense Object Detection.\n+# RetinaNet (Focal Loss for Dense Object Detection)\n## Model Zoo\n-| Backbone | Model | mstrain | imgs/GPU | lr schedule | FPS | Box AP | download | config |\n-| ------------ | --------- | ------- | -------- | ----------- | --- | ------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------- |\n-| ResNet50-FPN | RetinaNet | Yes | 4 | 1x | --- | 37.5 | [model](https://bj.bcebos.com/v1/paddledet/models/retinanet_r50_fpn_mstrain_1x_coco.pdparams)\\|[log](https://bj.bcebos.com/v1/paddledet/logs/retinanet_r50_fpn_mstrain_1x_coco.log) | retinanet_r50_fpn_mstrain_1x_coco.yml |\n-\n+| Backbone | Model | imgs/GPU | lr schedule | FPS | Box AP | download | config |\n+| ------------ | --------- | -------- | ----------- | --- | ------ | ---------- | ----------- |\n+| ResNet50-FPN | RetinaNet | 2 | 1x | --- | 37.5 | [model](https://bj.bcebos.com/v1/paddledet/models/retinanet_r50_fpn_1x_coco.pdparams) | [config](./retinanet_r50_fpn_1x_coco.yml) |\n**Notes:**\n-- All above models are trained on COCO train2017 with 4 GPUs and evaludated on val2017. Box AP=`mAP(IoU=0.5:0.95)`.\n+- All above models are trained on COCO train2017 with 8 GPUs and evaludated on val2017. Box AP=`mAP(IoU=0.5:0.95)`.\n-- Config `configs/retinanet/retinanet_r50_fpn_1x_coco.yml` is for 8 GPUs and `configs/retinanet/retinanet_r50_fpn_mstrain_1x_coco.yml` is for 4 GPUs (mind the difference of train batch size).\n## Citation\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/retinanet/_base_/retinanet_r50_fpn.yml",
"new_path": "configs/retinanet/_base_/retinanet_r50_fpn.yml",
"diff": "@@ -22,10 +22,6 @@ FPN:\nuse_c5: false\nRetinaHead:\n- num_classes: 80\n- prior_prob: 0.01\n- nms_pre: 1000\n- decode_reg_out: false\nconv_feat:\nname: RetinaFeat\nfeat_in: 256\n@@ -44,10 +40,6 @@ RetinaHead:\npositive_overlap: 0.5\nnegative_overlap: 0.4\nallow_low_quality: true\n- bbox_coder:\n- name: DeltaBBoxCoder\n- norm_mean: [0.0, 0.0, 0.0, 0.0]\n- norm_std: [1.0, 1.0, 1.0, 1.0]\nloss_class:\nname: FocalLoss\ngamma: 2.0\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/retinanet/_base_/retinanet_reader.yml",
"new_path": "configs/retinanet/_base_/retinanet_reader.yml",
"diff": "@@ -2,38 +2,35 @@ worker_num: 2\nTrainReader:\nsample_transforms:\n- Decode: {}\n- - RandomFlip: {prob: 0.5}\n- - Resize: {target_size: [800, 1333], keep_ratio: true, interp: 1}\n- - NormalizeImage: {is_scale: true, mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225]}\n+ - RandomResize: {target_size: [[640, 1333], [672, 1333], [704, 1333], [736, 1333], [768, 1333], [800, 1333]], keep_ratio: True, interp: 1}\n+ - RandomFlip: {}\n+ - NormalizeImage: {is_scale: True, mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225]}\n- Permute: {}\nbatch_transforms:\n- PadBatch: {pad_to_stride: 32}\nbatch_size: 2\n- shuffle: true\n- drop_last: true\n- use_process: true\n- collate_batch: false\n+ shuffle: True\n+ drop_last: True\n+ collate_batch: False\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: [800, 1333], keep_ratio: true, interp: 1}\n- - NormalizeImage: {is_scale: true, mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225]}\n+ - Resize: {target_size: [800, 1333], keep_ratio: True, interp: 1}\n+ - NormalizeImage: {is_scale: True, mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225]}\n- Permute: {}\nbatch_transforms:\n- PadBatch: {pad_to_stride: 32}\n- batch_size: 2\n- shuffle: false\n+ batch_size: 8\nTestReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: [800, 1333], keep_ratio: true, interp: 1}\n- - NormalizeImage: {is_scale: true, mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225]}\n+ - Resize: {target_size: [800, 1333], keep_ratio: True, interp: 1}\n+ - NormalizeImage: {is_scale: True, mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225]}\n- Permute: {}\nbatch_transforms:\n- PadBatch: {pad_to_stride: 32}\nbatch_size: 1\n- shuffle: false\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/retinanet/retinanet_r50_fpn_1x_coco.yml",
"new_path": "configs/retinanet/retinanet_r50_fpn_1x_coco.yml",
"diff": "@@ -7,4 +7,3 @@ _BASE_: [\n]\nweights: output/retinanet_r50_fpn_1x_coco/model_final\n-find_unused_parameters: true\n\\ No newline at end of file\n"
},
{
"change_type": "DELETE",
"old_path": "configs/retinanet/retinanet_r50_fpn_mstrain_1x_coco.yml",
"new_path": null,
"diff": "-_BASE_: [\n- '../datasets/coco_detection.yml',\n- '../runtime.yml',\n- '_base_/retinanet_r50_fpn.yml',\n- '_base_/optimizer_1x.yml',\n- '_base_/retinanet_reader.yml'\n-]\n-\n-worker_num: 4\n-TrainReader:\n- batch_size: 4\n- sample_transforms:\n- - Decode: {}\n- - RandomFlip: {prob: 0.5}\n- - RandomResize: {target_size: [[640, 1333], [672, 1333], [704, 1333], [736, 1333], [768, 1333], [800, 1333]], keep_ratio: true, interp: 1}\n- - NormalizeImage: {is_scale: true, mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225]}\n- - Permute: {}\n-\n-weights: output/retinanet_r50_fpn_mstrain_1x_coco/model_final\n-find_unused_parameters: true\n\\ No newline at end of file\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/__init__.py",
"new_path": "ppdet/modeling/__init__.py",
"diff": "@@ -29,7 +29,6 @@ from . import reid\nfrom . import mot\nfrom . import transformers\nfrom . import assigners\n-from . import coders\nfrom .ops import *\nfrom .backbones import *\n@@ -44,4 +43,3 @@ from .reid import *\nfrom .mot import *\nfrom .transformers import *\nfrom .assigners import *\n-from .coders import *\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/architectures/retinanet.py",
"new_path": "ppdet/modeling/architectures/retinanet.py",
"diff": "@@ -22,14 +22,12 @@ import paddle\n__all__ = ['RetinaNet']\n+\n@register\nclass RetinaNet(BaseArch):\n__category__ = 'architecture'\n- def __init__(self,\n- backbone,\n- neck,\n- head):\n+ def __init__(self, backbone, neck, head):\nsuper(RetinaNet, self).__init__()\nself.backbone = backbone\nself.neck = neck\n@@ -38,35 +36,33 @@ class RetinaNet(BaseArch):\n@classmethod\ndef from_config(cls, cfg, *args, **kwargs):\nbackbone = create(cfg['backbone'])\n+\nkwargs = {'input_shape': backbone.out_shape}\nneck = create(cfg['neck'], **kwargs)\n- head = create(cfg['head'])\n+\n+ kwargs = {'input_shape': neck.out_shape}\n+ head = create(cfg['head'], **kwargs)\n+\nreturn {\n'backbone': backbone,\n'neck': neck,\n- 'head': head}\n+ 'head': head,\n+ }\ndef _forward(self):\nbody_feats = self.backbone(self.inputs)\nneck_feats = self.neck(body_feats)\n+\n+ if self.training:\n+ return self.head(neck_feats, self.inputs)\n+ else:\nhead_outs = self.head(neck_feats)\n- if not self.training:\n- im_shape = self.inputs['im_shape']\n- scale_factor = self.inputs['scale_factor']\n- bboxes, bbox_num = self.head.post_process(head_outs, im_shape, scale_factor)\n- return bboxes, bbox_num\n- return head_outs\n+ bbox, bbox_num = self.head.post_process(\n+ head_outs, self.inputs['im_shape'], self.inputs['scale_factor'])\n+ return {'bbox': bbox, 'bbox_num': bbox_num}\ndef get_loss(self):\n- loss = dict()\n- head_outs = self._forward()\n- loss_retina = self.head.get_loss(head_outs, self.inputs)\n- loss.update(loss_retina)\n- total_loss = paddle.add_n(list(loss.values()))\n- loss.update(loss=total_loss)\n- return loss\n+ return self._forward()\ndef get_pred(self):\n- bbox_pred, bbox_num = self._forward()\n- output = dict(bbox=bbox_pred, bbox_num=bbox_num)\n- return output\n+ return self._forward()\n"
},
{
"change_type": "DELETE",
"old_path": "ppdet/modeling/coders/__init__.py",
"new_path": null,
"diff": "-from .delta_bbox_coder import DeltaBBoxCoder\n"
},
{
"change_type": "DELETE",
"old_path": "ppdet/modeling/coders/delta_bbox_coder.py",
"new_path": null,
"diff": "-import paddle\n-import numpy as np\n-from ppdet.core.workspace import register\n-from ppdet.modeling.bbox_utils import delta2bbox_v2, bbox2delta_v2\n-\n-__all__ = ['DeltaBBoxCoder']\n-\n-\n-@register\n-class DeltaBBoxCoder:\n- \"\"\"Encode bboxes in terms of delta/offset of a reference bbox.\n- Args:\n- norm_mean (list[float]): the mean to normalize delta\n- norm_std (list[float]): the std to normalize delta\n- wh_ratio_clip (float): to clip delta wh of decoded bboxes\n- ctr_clip (float or None): whether to clip delta xy of decoded bboxes\n- \"\"\"\n- def __init__(self,\n- norm_mean=[0.0, 0.0, 0.0, 0.0],\n- norm_std=[1., 1., 1., 1.],\n- wh_ratio_clip=16/1000.0,\n- ctr_clip=None):\n- self.norm_mean = norm_mean\n- self.norm_std = norm_std\n- self.wh_ratio_clip = wh_ratio_clip\n- self.ctr_clip = ctr_clip\n-\n- def encode(self, bboxes, tar_bboxes):\n- return bbox2delta_v2(\n- bboxes, tar_bboxes, means=self.norm_mean, stds=self.norm_std)\n-\n- def decode(self, bboxes, deltas, max_shape=None):\n- return delta2bbox_v2(\n- bboxes,\n- deltas,\n- max_shape=max_shape,\n- wh_ratio_clip=self.wh_ratio_clip,\n- ctr_clip=self.ctr_clip,\n- means=self.norm_mean,\n- stds=self.norm_std)\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/retina_head.py",
"new_path": "ppdet/modeling/heads/retina_head.py",
"diff": "-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.\n+# Copyright (c) 2021 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@@ -16,17 +16,20 @@ from __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n-import math, paddle\n+import math\n+import paddle\nimport paddle.nn as nn\nimport paddle.nn.functional as F\nfrom paddle import ParamAttr\nfrom paddle.nn.initializer import Normal, Constant\n-from ppdet.modeling.proposal_generator import AnchorGenerator\n-from ppdet.core.workspace import register\n+from ppdet.modeling.bbox_utils import bbox2delta, delta2bbox\nfrom ppdet.modeling.heads.fcos_head import FCOSFeat\n+from ppdet.core.workspace import register\n+\n__all__ = ['RetinaHead']\n+\n@register\nclass RetinaFeat(FCOSFeat):\n\"\"\"We use FCOSFeat to construct conv layers in RetinaNet.\n@@ -34,72 +37,49 @@ class RetinaFeat(FCOSFeat):\n\"\"\"\npass\n-@register\n-class RetinaAnchorGenerator(AnchorGenerator):\n- def __init__(self,\n- octave_base_scale=4,\n- scales_per_octave=3,\n- aspect_ratios=[0.5, 1.0, 2.0],\n- strides=[8.0, 16.0, 32.0, 64.0, 128.0],\n- variance=[1.0, 1.0, 1.0, 1.0],\n- offset=0.0):\n- anchor_sizes = []\n- for s in strides:\n- anchor_sizes.append([\n- s * octave_base_scale * 2**(i/scales_per_octave) \\\n- for i in range(scales_per_octave)])\n- super(RetinaAnchorGenerator, self).__init__(\n- anchor_sizes=anchor_sizes,\n- aspect_ratios=aspect_ratios,\n- strides=strides,\n- variance=variance,\n- offset=offset)\n@register\nclass RetinaHead(nn.Layer):\n\"\"\"Used in RetinaNet proposed in paper https://arxiv.org/pdf/1708.02002.pdf\n\"\"\"\n+ __shared__ = ['num_classes']\n__inject__ = [\n- 'conv_feat', 'anchor_generator', 'bbox_assigner',\n- 'bbox_coder', 'loss_class', 'loss_bbox', 'nms']\n+ 'conv_feat', 'anchor_generator', 'bbox_assigner', 'loss_class',\n+ 'loss_bbox', 'nms'\n+ ]\n+\ndef __init__(self,\nnum_classes=80,\n+ conv_feat='RetinaFeat',\n+ anchor_generator='RetinaAnchorGenerator',\n+ bbox_assigner='MaxIoUAssigner',\n+ loss_class='FocalLoss',\n+ loss_bbox='SmoothL1Loss',\n+ nms='MultiClassNMS',\nprior_prob=0.01,\n- decode_reg_out=False,\n- conv_feat=None,\n- anchor_generator=None,\n- bbox_assigner=None,\n- bbox_coder=None,\n- loss_class=None,\n- loss_bbox=None,\nnms_pre=1000,\n- nms=None):\n+ weights=[1., 1., 1., 1.]):\nsuper(RetinaHead, self).__init__()\nself.num_classes = num_classes\n- self.prior_prob = prior_prob\n- # allow RetinaNet to use IoU based losses.\n- self.decode_reg_out = decode_reg_out\nself.conv_feat = conv_feat\nself.anchor_generator = anchor_generator\nself.bbox_assigner = bbox_assigner\n- self.bbox_coder = bbox_coder\nself.loss_class = loss_class\nself.loss_bbox = loss_bbox\n- self.nms_pre = nms_pre\nself.nms = nms\n- self.cls_out_channels = num_classes\n- self.init_layers()\n+ self.nms_pre = nms_pre\n+ self.weights = weights\n- def init_layers(self):\n- bias_init_value = -math.log((1 - self.prior_prob) / self.prior_prob)\n+ bias_init_value = -math.log((1 - prior_prob) / prior_prob)\nnum_anchors = self.anchor_generator.num_anchors\nself.retina_cls = nn.Conv2D(\nin_channels=self.conv_feat.feat_out,\n- out_channels=self.cls_out_channels * num_anchors,\n+ out_channels=self.num_classes * num_anchors,\nkernel_size=3,\nstride=1,\npadding=1,\n- weight_attr=ParamAttr(initializer=Normal(mean=0.0, std=0.01)),\n+ weight_attr=ParamAttr(initializer=Normal(\n+ mean=0.0, std=0.01)),\nbias_attr=ParamAttr(initializer=Constant(value=bias_init_value)))\nself.retina_reg = nn.Conv2D(\nin_channels=self.conv_feat.feat_out,\n@@ -107,10 +87,11 @@ class RetinaHead(nn.Layer):\nkernel_size=3,\nstride=1,\npadding=1,\n- weight_attr=ParamAttr(initializer=Normal(mean=0.0, std=0.01)),\n+ weight_attr=ParamAttr(initializer=Normal(\n+ mean=0.0, std=0.01)),\nbias_attr=ParamAttr(initializer=Constant(value=0)))\n- def forward(self, neck_feats):\n+ def forward(self, neck_feats, targets=None):\ncls_logits_list = []\nbboxes_reg_list = []\nfor neck_feat in neck_feats:\n@@ -119,33 +100,40 @@ class RetinaHead(nn.Layer):\nbbox_reg = self.retina_reg(conv_reg_feat)\ncls_logits_list.append(cls_logits)\nbboxes_reg_list.append(bbox_reg)\n- return (cls_logits_list, bboxes_reg_list)\n- def get_loss(self, head_outputs, meta):\n+ if self.training:\n+ return self.get_loss([cls_logits_list, bboxes_reg_list], targets)\n+ else:\n+ return [cls_logits_list, bboxes_reg_list]\n+\n+ def get_loss(self, head_outputs, targets):\n\"\"\"Here we calculate loss for a batch of images.\nWe assign anchors to gts in each image and gather all the assigned\npostive and negative samples. Then loss is calculated on the gathered\nsamples.\n\"\"\"\n- cls_logits, bboxes_reg = head_outputs\n- # we use the same anchor for all images\n- anchors = self.anchor_generator(cls_logits)\n+ cls_logits_list, bboxes_reg_list = head_outputs\n+ anchors = self.anchor_generator(cls_logits_list)\nanchors = paddle.concat(anchors)\n# matches: contain gt_inds\n# match_labels: -1(ignore), 0(neg) or 1(pos)\nmatches_list, match_labels_list = [], []\n# assign anchors to gts, no sampling is involved\n- for gt_bbox in meta['gt_bbox']:\n+ for gt_bbox in targets['gt_bbox']:\nmatches, match_labels = self.bbox_assigner(anchors, gt_bbox)\nmatches_list.append(matches)\nmatch_labels_list.append(match_labels)\n+\n# reshape network outputs\n- cls_logits = [_.transpose([0, 2, 3, 1]) for _ in cls_logits]\n- cls_logits = [_.reshape([0, -1, self.cls_out_channels]) \\\n- for _ in cls_logits]\n- bboxes_reg = [_.transpose([0, 2, 3, 1]) for _ in bboxes_reg]\n- bboxes_reg = [_.reshape([0, -1, 4]) for _ in bboxes_reg]\n+ cls_logits = [\n+ _.transpose([0, 2, 3, 1]).reshape([0, -1, self.num_classes])\n+ for _ in cls_logits_list\n+ ]\n+ bboxes_reg = [\n+ _.transpose([0, 2, 3, 1]).reshape([0, -1, 4])\n+ for _ in bboxes_reg_list\n+ ]\ncls_logits = paddle.concat(cls_logits, axis=1)\nbboxes_reg = paddle.concat(bboxes_reg, axis=1)\n@@ -154,7 +142,7 @@ class RetinaHead(nn.Layer):\n# find and gather preds and targets in each image\nfor matches, match_labels, cls_logit, bbox_reg, gt_bbox, gt_class in \\\nzip(matches_list, match_labels_list, cls_logits, bboxes_reg,\n- meta['gt_bbox'], meta['gt_class']):\n+ targets['gt_bbox'], targets['gt_class']):\npos_mask = (match_labels == 1)\nneg_mask = (match_labels == 0)\nchosen_mask = paddle.logical_or(pos_mask, neg_mask)\n@@ -163,19 +151,16 @@ class RetinaHead(nn.Layer):\nbg_class = paddle.to_tensor(\n[self.num_classes], dtype=gt_class.dtype)\n# a trick to assign num_classes to negative targets\n- gt_class = paddle.concat([gt_class, bg_class])\n- matches = paddle.where(\n- neg_mask, paddle.full_like(matches, gt_class.size-1), matches)\n+ gt_class = paddle.concat([gt_class, bg_class], axis=-1)\n+ matches = paddle.where(neg_mask,\n+ paddle.full_like(matches, gt_class.size - 1),\n+ matches)\ncls_pred = cls_logit[chosen_mask]\ncls_tar = gt_class[matches[chosen_mask]]\nreg_pred = bbox_reg[pos_mask].reshape([-1, 4])\nreg_tar = gt_bbox[matches[pos_mask]].reshape([-1, 4])\n- if self.decode_reg_out:\n- reg_pred = self.bbox_coder.decode(\n- anchors[pos_mask], reg_pred)\n- else:\n- reg_tar = self.bbox_coder.encode(anchors[pos_mask], reg_tar)\n+ reg_tar = bbox2delta(anchors[pos_mask], reg_tar, self.weights)\ncls_pred_list.append(cls_pred)\ncls_tar_list.append(cls_tar)\nreg_pred_list.append(reg_pred)\n@@ -184,27 +169,38 @@ class RetinaHead(nn.Layer):\ncls_tar = paddle.concat(cls_tar_list)\nreg_pred = paddle.concat(reg_pred_list)\nreg_tar = paddle.concat(reg_tar_list)\n+\navg_factor = max(1.0, reg_pred.shape[0])\ncls_loss = self.loss_class(\ncls_pred, cls_tar, reduction='sum') / avg_factor\n- if reg_pred.size == 0:\n- reg_loss = bboxes_reg[0][0].sum() * 0\n+\n+ if reg_pred.shape[0] == 0:\n+ reg_loss = paddle.zeros([1])\n+ reg_loss.stop_gradient = False\nelse:\nreg_loss = self.loss_bbox(\nreg_pred, reg_tar, reduction='sum') / avg_factor\n- return dict(loss_cls=cls_loss, loss_reg=reg_loss)\n+\n+ loss = cls_loss + reg_loss\n+ out_dict = {\n+ 'loss_cls': cls_loss,\n+ 'loss_reg': reg_loss,\n+ 'loss': loss,\n+ }\n+ return out_dict\ndef get_bboxes_single(self,\nanchors,\n- cls_scores,\n- bbox_preds,\n+ cls_scores_list,\n+ bbox_preds_list,\nim_shape,\nscale_factor,\nrescale=True):\n- assert len(cls_scores) == len(bbox_preds)\n+ assert len(cls_scores_list) == len(bbox_preds_list)\nmlvl_bboxes = []\nmlvl_scores = []\n- for anchor, cls_score, bbox_pred in zip(anchors, cls_scores, bbox_preds):\n+ for anchor, cls_score, bbox_pred in zip(anchors, cls_scores_list,\n+ bbox_preds_list):\ncls_score = cls_score.reshape([-1, self.num_classes])\nbbox_pred = bbox_pred.reshape([-1, 4])\nif self.nms_pre is not None and cls_score.shape[0] > self.nms_pre:\n@@ -213,9 +209,7 @@ class RetinaHead(nn.Layer):\nbbox_pred = bbox_pred.gather(topk_inds)\nanchor = anchor.gather(topk_inds)\ncls_score = cls_score.gather(topk_inds)\n- bbox_pred = self.bbox_coder.decode(\n- anchor, bbox_pred, max_shape=im_shape)\n- bbox_pred = bbox_pred.squeeze()\n+ bbox_pred = delta2bbox(bbox_pred, anchor, self.weights).squeeze()\nmlvl_bboxes.append(bbox_pred)\nmlvl_scores.append(F.sigmoid(cls_score))\nmlvl_bboxes = paddle.concat(mlvl_bboxes)\n@@ -227,18 +221,15 @@ class RetinaHead(nn.Layer):\nmlvl_scores = mlvl_scores.transpose([1, 0])\nreturn mlvl_bboxes, mlvl_scores\n- def decode(self, anchors, cls_scores, bbox_preds, im_shape, scale_factor):\n+ def decode(self, anchors, cls_logits, bboxes_reg, im_shape, scale_factor):\nbatch_bboxes = []\nbatch_scores = []\n- for img_id in range(cls_scores[0].shape[0]):\n- num_lvls = len(cls_scores)\n- cls_score_list = [cls_scores[i][img_id] for i in range(num_lvls)]\n- bbox_pred_list = [bbox_preds[i][img_id] for i in range(num_lvls)]\n+ for img_id in range(cls_logits[0].shape[0]):\n+ num_lvls = len(cls_logits)\n+ cls_scores_list = [cls_logits[i][img_id] for i in range(num_lvls)]\n+ bbox_preds_list = [bboxes_reg[i][img_id] for i in range(num_lvls)]\nbboxes, scores = self.get_bboxes_single(\n- anchors,\n- cls_score_list,\n- bbox_pred_list,\n- im_shape[img_id],\n+ anchors, cls_scores_list, bbox_preds_list, im_shape[img_id],\nscale_factor[img_id])\nbatch_bboxes.append(bboxes)\nbatch_scores.append(scores)\n@@ -247,11 +238,12 @@ class RetinaHead(nn.Layer):\nreturn batch_bboxes, batch_scores\ndef post_process(self, head_outputs, im_shape, scale_factor):\n- cls_scores, bbox_preds = head_outputs\n- anchors = self.anchor_generator(cls_scores)\n- cls_scores = [_.transpose([0, 2, 3, 1]) for _ in cls_scores]\n- bbox_preds = [_.transpose([0, 2, 3, 1]) for _ in bbox_preds]\n- bboxes, scores = self.decode(\n- anchors, cls_scores, bbox_preds, im_shape, scale_factor)\n+ cls_logits_list, bboxes_reg_list = head_outputs\n+ anchors = self.anchor_generator(cls_logits_list)\n+ cls_logits = [_.transpose([0, 2, 3, 1]) for _ in cls_logits_list]\n+ bboxes_reg = [_.transpose([0, 2, 3, 1]) for _ in bboxes_reg_list]\n+ bboxes, scores = self.decode(anchors, cls_logits, bboxes_reg, im_shape,\n+ scale_factor)\n+\nbbox_pred, bbox_num, _ = self.nms(bboxes, scores)\nreturn bbox_pred, bbox_num\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/proposal_generator/anchor_generator.py",
"new_path": "ppdet/modeling/proposal_generator/anchor_generator.py",
"diff": "@@ -22,6 +22,8 @@ import paddle.nn as nn\nfrom ppdet.core.workspace import register\n+__all__ = ['AnchorGenerator', 'RetinaAnchorGenerator']\n+\n@register\nclass AnchorGenerator(nn.Layer):\n@@ -129,3 +131,25 @@ class AnchorGenerator(nn.Layer):\nFor FPN models, `num_anchors` on every feature map is the same.\n\"\"\"\nreturn len(self.cell_anchors[0])\n+\n+\n+@register\n+class RetinaAnchorGenerator(AnchorGenerator):\n+ def __init__(self,\n+ octave_base_scale=4,\n+ scales_per_octave=3,\n+ aspect_ratios=[0.5, 1.0, 2.0],\n+ strides=[8.0, 16.0, 32.0, 64.0, 128.0],\n+ variance=[1.0, 1.0, 1.0, 1.0],\n+ offset=0.0):\n+ anchor_sizes = []\n+ for s in strides:\n+ anchor_sizes.append([\n+ s * octave_base_scale * 2**(i/scales_per_octave) \\\n+ for i in range(scales_per_octave)])\n+ super(RetinaAnchorGenerator, self).__init__(\n+ anchor_sizes=anchor_sizes,\n+ aspect_ratios=aspect_ratios,\n+ strides=strides,\n+ variance=variance,\n+ offset=offset)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
refine RetinaNet codes (#5796)
|
499,304 |
22.04.2022 11:43:04
| -28,800 |
ad0a436a12d45ec3cd13ae16bac79611908a872b
|
fix AugmentHSV output type
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/transform/operators.py",
"new_path": "ppdet/data/transform/operators.py",
"diff": "@@ -2202,7 +2202,7 @@ class AugmentHSV(BaseOperator):\nelse:\ncv2.cvtColor(img_hsv, cv2.COLOR_HSV2RGB, dst=img)\n- sample['image'] = img\n+ sample['image'] = img.astype(np.float32)\nreturn sample\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix AugmentHSV output type (#5800)
|
499,298 |
22.04.2022 15:23:31
| -28,800 |
7c5736b2d46c940caf4b6a35c6abfa533bfb9ff6
|
fix yolox cpp infer
|
[
{
"change_type": "MODIFY",
"old_path": "configs/yolox/_base_/yolox_reader.yml",
"new_path": "configs/yolox/_base_/yolox_reader.yml",
"diff": "@@ -27,8 +27,8 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: 640, keep_ratio: True}\n- - Pad: {size: 640, fill_value: [114., 114., 114.]}\n+ - Resize: {target_size: [640, 640], keep_ratio: True, interp: 1}\n+ - Pad: {size: [640, 640], fill_value: [114., 114., 114.]}\n- Permute: {}\nbatch_size: 4\n@@ -38,7 +38,7 @@ TestReader:\nimage_shape: [3, 640, 640]\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: 640, keep_ratio: True}\n- - Pad: {size: 640, fill_value: [114., 114., 114.]}\n+ - Resize: {target_size: [640, 640], keep_ratio: True, interp: 1}\n+ - Pad: {size: [640, 640], fill_value: [114., 114., 114.]}\n- Permute: {}\nbatch_size: 1\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/yolox/yolox_nano_300e_coco.yml",
"new_path": "configs/yolox/yolox_nano_300e_coco.yml",
"diff": "@@ -64,8 +64,8 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: 416, keep_ratio: True}\n- - Pad: {size: 416, fill_value: [114., 114., 114.]}\n+ - Resize: {target_size: [416, 416], keep_ratio: True, interp: 1}\n+ - Pad: {size: [416, 416], fill_value: [114., 114., 114.]}\n- Permute: {}\nbatch_size: 8\n@@ -75,7 +75,7 @@ TestReader:\nimage_shape: [3, 416, 416]\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: 416, keep_ratio: True}\n- - Pad: {size: 416, fill_value: [114., 114., 114.]}\n+ - Resize: {target_size: [416, 416], keep_ratio: True, interp: 1}\n+ - Pad: {size: [416, 416], fill_value: [114., 114., 114.]}\n- Permute: {}\nbatch_size: 1\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/yolox/yolox_tiny_300e_coco.yml",
"new_path": "configs/yolox/yolox_tiny_300e_coco.yml",
"diff": "@@ -52,8 +52,8 @@ TrainReader:\nEvalReader:\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: 416, keep_ratio: True}\n- - Pad: {size: 416, fill_value: [114., 114., 114.]}\n+ - Resize: {target_size: [416, 416], keep_ratio: True, interp: 1}\n+ - Pad: {size: [416, 416], fill_value: [114., 114., 114.]}\n- Permute: {}\nbatch_size: 8\n@@ -63,7 +63,7 @@ TestReader:\nimage_shape: [3, 416, 416]\nsample_transforms:\n- Decode: {}\n- - Resize: {target_size: 416, keep_ratio: True}\n- - Pad: {size: 416, fill_value: [114., 114., 114.]}\n+ - Resize: {target_size: [416, 416], keep_ratio: True, interp: 1}\n+ - Pad: {size: [416, 416], fill_value: [114., 114., 114.]}\n- Permute: {}\nbatch_size: 1\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/include/preprocess_op.h",
"new_path": "deploy/cpp/include/preprocess_op.h",
"diff": "@@ -161,6 +161,20 @@ class WarpAffine : public PreprocessOp {\nint pad_ = 31;\n};\n+class Pad : public PreprocessOp {\n+ public:\n+ virtual void Init(const YAML::Node& item) {\n+ size_ = item[\"size\"].as<std::vector<int>>();\n+ fill_value_ = item[\"fill_value\"].as<std::vector<float>>();\n+ }\n+\n+ virtual void Run(cv::Mat* im, ImageBlob* data);\n+\n+ private:\n+ std::vector<int> size_;\n+ std::vector<float> fill_value_;\n+};\n+\nvoid CropImg(cv::Mat& img,\ncv::Mat& crop_img,\nstd::vector<int>& area,\n@@ -203,6 +217,8 @@ class Preprocessor {\nreturn std::make_shared<TopDownEvalAffine>();\n} else if (name == \"WarpAffine\") {\nreturn std::make_shared<WarpAffine>();\n+ }else if (name == \"Pad\") {\n+ return std::make_shared<Pad>();\n}\nstd::cerr << \"can not find function of OP: \" << name\n<< \" and return: nullptr\" << std::endl;\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/src/preprocess_op.cc",
"new_path": "deploy/cpp/src/preprocess_op.cc",
"diff": "@@ -229,6 +229,23 @@ void WarpAffine::Run(cv::Mat* im, ImageBlob* data) {\n};\n}\n+void Pad::Run(cv::Mat* im, ImageBlob* data) {\n+ int h = size_[0];\n+ int w = size_[1];\n+ int rh = im->rows;\n+ int rw = im->cols;\n+ if (h == rh && w == rw){\n+ data->in_net_im_ = im->clone();\n+ return;\n+ }\n+ cv::copyMakeBorder(\n+ *im, *im, 0, h - rh, 0, w - rw, cv::BORDER_CONSTANT, cv::Scalar(114));\n+ data->in_net_im_ = im->clone();\n+ data->in_net_shape_ = {\n+ static_cast<float>(im->rows), static_cast<float>(im->cols),\n+ };\n+}\n+\n// Preprocessor op running order\nconst std::vector<std::string> Preprocessor::RUN_ORDER = {\"InitInfo\",\n\"TopDownEvalAffine\",\n@@ -237,7 +254,8 @@ const std::vector<std::string> Preprocessor::RUN_ORDER = {\"InitInfo\",\n\"WarpAffine\",\n\"NormalizeImage\",\n\"PadStride\",\n- \"Permute\"};\n+ \"Permute\",\n+ \"Pad\",};\nvoid Preprocessor::Run(cv::Mat* im, ImageBlob* data) {\nfor (const auto& name : RUN_ORDER) {\n"
},
{
"change_type": "MODIFY",
"old_path": "deploy/python/preprocess.py",
"new_path": "deploy/python/preprocess.py",
"diff": "@@ -247,77 +247,30 @@ class LetterBoxResize(object):\nclass Pad(object):\n- def __init__(self,\n- size=None,\n- size_divisor=32,\n- pad_mode=0,\n- offsets=None,\n- fill_value=(127.5, 127.5, 127.5)):\n+ def __init__(self, size, fill_value=[114.0, 114.0, 114.0]):\n\"\"\"\n- Pad image to a specified size or multiple of size_divisor.\n+ Pad image to a specified size.\nArgs:\n- size (int, Sequence): image target size, if None, pad to multiple of size_divisor, default None\n- size_divisor (int): size divisor, default 32\n- pad_mode (int): pad mode, currently only supports four modes [-1, 0, 1, 2]. if -1, use specified offsets\n- if 0, only pad to right and bottom. if 1, pad according to center. if 2, only pad left and top\n- offsets (list): [offset_x, offset_y], specify offset while padding, only supported pad_mode=-1\n- fill_value (bool): rgb value of pad area, default (127.5, 127.5, 127.5)\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\"\"\"\nsuper(Pad, self).__init__()\nif isinstance(size, int):\nsize = [size, size]\n-\n- assert pad_mode in [\n- -1, 0, 1, 2\n- ], 'currently only supports four modes [-1, 0, 1, 2]'\n- if pad_mode == -1:\n- assert offsets, 'if pad_mode is -1, offsets should not be None'\n-\nself.size = size\n- self.size_divisor = size_divisor\n- self.pad_mode = pad_mode\nself.fill_value = fill_value\n- self.offsets = offsets\n-\n- def apply_image(self, image, offsets, im_size, size):\n- x, y = offsets\n- im_h, im_w = im_size\n- h, w = size\n- canvas = np.ones((h, w, 3), dtype=np.float32)\n- canvas *= np.array(self.fill_value, dtype=np.float32)\n- canvas[y:y + im_h, x:x + im_w, :] = image.astype(np.float32)\n- return canvas\ndef __call__(self, im, im_info):\nim_h, im_w = im.shape[:2]\n- if self.size:\nh, w = self.size\n- assert (\n- im_h <= h and im_w <= w\n- ), '(h, w) of target size should be greater than (im_h, im_w)'\n- else:\n- h = int(np.ceil(im_h / self.size_divisor) * self.size_divisor)\n- w = int(np.ceil(im_w / self.size_divisor) * self.size_divisor)\n-\nif h == im_h and w == im_w:\nim = im.astype(np.float32)\nreturn im, im_info\n- if self.pad_mode == -1:\n- offset_x, offset_y = self.offsets\n- elif self.pad_mode == 0:\n- offset_y, offset_x = 0, 0\n- elif self.pad_mode == 1:\n- offset_y, offset_x = (h - im_h) // 2, (w - im_w) // 2\n- else:\n- offset_y, offset_x = h - im_h, w - im_w\n-\n- offsets, im_size, size = [offset_x, offset_y], [im_h, im_w], [h, w]\n- im = self.apply_image(im, offsets, im_size, size)\n-\n- if self.pad_mode == 0:\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\nreturn im, im_info\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix yolox cpp infer (#5804)
|
499,304 |
22.04.2022 19:40:36
| -28,800 |
97e9e75bd1c746084d8f047cfd7abf8cc64c9263
|
fix transform resize type to float32
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/data/transform/operators.py",
"new_path": "ppdet/data/transform/operators.py",
"diff": "@@ -824,7 +824,7 @@ class Resize(BaseOperator):\nim_scale_x = resize_w / im_shape[1]\nim = self.apply_image(sample['image'], [im_scale_x, im_scale_y])\n- sample['image'] = im\n+ sample['image'] = im.astype(np.float32)\nsample['im_shape'] = np.asarray([resize_h, resize_w], dtype=np.float32)\nif 'scale_factor' in sample:\nscale_factor = sample['scale_factor']\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix transform resize type to float32 (#5802)
|
499,386 |
24.04.2022 07:54:05
| 18,000 |
8bdcd15723d735de8a4f36816d4a94fc27003673
|
fix dark postprocess in cpp infer deploy(
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/src/keypoint_postprocess.cc",
"new_path": "deploy/cpp/src/keypoint_postprocess.cc",
"diff": "@@ -193,7 +193,7 @@ void get_final_preds(std::vector<float>& heatmap,\nint px = int(coords[j * 2] + 0.5);\nint py = int(coords[j * 2 + 1] + 0.5);\n- if(DARK && px > 1 && px < heatmap_width - 2){\n+ if(DARK && px > 1 && px < heatmap_width - 2 && py > 1 && py < heatmap_height - 2){\ndark_parse(heatmap, dim, coords, px, py, index, j);\n}\nelse{\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix dark postprocess in cpp infer deploy( #5751) (#5815)
|
499,333 |
28.04.2022 11:55:54
| -28,800 |
9535ca879fc01b38ab39659370909caf5583ebcf
|
yaml support no constructor
|
[
{
"change_type": "MODIFY",
"old_path": "configs/datasets/roadsign_voc.yml",
"new_path": "configs/datasets/roadsign_voc.yml",
"diff": "@@ -3,19 +3,19 @@ map_type: integral\nnum_classes: 4\nTrainDataset:\n- !VOCDataSet\n+ name: VOCDataSet\ndataset_dir: dataset/roadsign_voc\nanno_path: train.txt\nlabel_list: label_list.txt\ndata_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']\nEvalDataset:\n- !VOCDataSet\n+ name: VOCDataSet\ndataset_dir: dataset/roadsign_voc\nanno_path: valid.txt\nlabel_list: label_list.txt\ndata_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']\nTestDataset:\n- !ImageFolder\n+ name: ImageFolder\nanno_path: dataset/roadsign_voc/label_list.txt\n"
},
{
"change_type": "MODIFY",
"old_path": "configs/yolov3/_base_/optimizer_40e.yml",
"new_path": "configs/yolov3/_base_/optimizer_40e.yml",
"diff": "@@ -3,12 +3,12 @@ epoch: 40\nLearningRate:\nbase_lr: 0.0001\nschedulers:\n- - !PiecewiseDecay\n+ - name: PiecewiseDecay\ngamma: 0.1\nmilestones:\n- 32\n- 36\n- - !LinearWarmup\n+ - name: LinearWarmup\nstart_factor: 0.3333333333333333\nsteps: 100\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/core/workspace.py",
"new_path": "ppdet/core/workspace.py",
"diff": "@@ -210,9 +210,17 @@ def create(cls_or_name, **kwargs):\nassert type(cls_or_name) in [type, str\n], \"should be a class or name of a class\"\nname = type(cls_or_name) == str and cls_or_name or cls_or_name.__name__\n- assert name in global_config and \\\n- isinstance(global_config[name], SchemaDict), \\\n- \"the module {} is not registered\".format(name)\n+ if name in global_config:\n+ if isinstance(global_config[name], SchemaDict):\n+ pass\n+ elif hasattr(global_config[name], \"__dict__\"):\n+ # support instance return directly\n+ return global_config[name]\n+ else:\n+ raise ValueError(\"The module {} is not registered\".format(name))\n+ else:\n+ raise ValueError(\"The module {} is not registered\".format(name))\n+\nconfig = global_config[name]\ncls = getattr(config.pymodule, name)\ncls_kwargs = {}\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/data/source/dataset.py",
"new_path": "ppdet/data/source/dataset.py",
"diff": "@@ -23,6 +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@serializable\n@@ -60,6 +61,9 @@ class DetDataset(Dataset):\ndef __len__(self, ):\nreturn len(self.roidbs)\n+ def __call__(self, *args, **kwargs):\n+ return self\n+\ndef __getitem__(self, idx):\n# data batch\nroidb = copy.deepcopy(self.roidbs[idx])\n@@ -198,3 +202,40 @@ class ImageFolder(DetDataset):\ndef set_images(self, images):\nself.image_dir = images\nself.roidbs = self._load_images()\n+\n+\n+@register\n+class CommonDataset(object):\n+ def __init__(self, **dataset_args):\n+ super(CommonDataset, self).__init__()\n+ dataset_args = copy.deepcopy(dataset_args)\n+ type = dataset_args.pop(\"name\")\n+ self.dataset = getattr(source, type)(**dataset_args)\n+\n+ def __call__(self):\n+ return self.dataset\n+\n+\n+@register\n+class TrainDataset(CommonDataset):\n+ pass\n+\n+\n+@register\n+class EvalMOTDataset(CommonDataset):\n+ pass\n+\n+\n+@register\n+class TestMOTDataset(CommonDataset):\n+ pass\n+\n+\n+@register\n+class EvalDataset(CommonDataset):\n+ pass\n+\n+\n+@register\n+class TestDataset(CommonDataset):\n+ pass\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -68,9 +68,10 @@ class Trainer(object):\n# build data loader\nif cfg.architecture in MOT_ARCH and self.mode in ['eval', 'test']:\n- self.dataset = cfg['{}MOTDataset'.format(self.mode.capitalize())]\n+ self.dataset = create('{}MOTDataset'.format(self.mode.capitalize(\n+ )))()\nelse:\n- self.dataset = cfg['{}Dataset'.format(self.mode.capitalize())]\n+ self.dataset = create('{}Dataset'.format(self.mode.capitalize()))()\nif cfg.architecture == 'DeepSORT' and self.mode == 'train':\nlogger.error('DeepSORT has no need of training on mot dataset.')\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/optimizer.py",
"new_path": "ppdet/optimizer.py",
"diff": "@@ -16,6 +16,7 @@ from __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n+import sys\nimport math\nimport weakref\nimport paddle\n@@ -25,6 +26,7 @@ import paddle.optimizer as optimizer\nimport paddle.regularizer as regularizer\nfrom ppdet.core.workspace import register, serializable\n+import copy\n__all__ = ['LearningRate', 'OptimizerBuilder']\n@@ -252,7 +254,18 @@ class LearningRate(object):\nschedulers=[PiecewiseDecay(), LinearWarmup()]):\nsuper(LearningRate, self).__init__()\nself.base_lr = base_lr\n- self.schedulers = schedulers\n+ self.schedulers = []\n+\n+ schedulers = copy.deepcopy(schedulers)\n+ for sched in schedulers:\n+ if isinstance(sched, dict):\n+ # support dict sched instantiate\n+ module = sys.modules[__name__]\n+ type = sched.pop(\"name\")\n+ scheduler = getattr(module, type)(**sched)\n+ self.schedulers.append(scheduler)\n+ else:\n+ self.schedulers.append(sched)\ndef __call__(self, step_per_epoch):\nassert len(self.schedulers) >= 1\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
yaml support no constructor (#5847)
|
499,319 |
29.04.2022 11:28:49
| -28,800 |
81d28246b487580ceacaed4f16b98109e46731eb
|
[test_tipc] add paddle2onnx; fix onnxruntime infer_demo
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/third_engine/demo_onnxruntime/infer_demo.py",
"new_path": "deploy/third_engine/demo_onnxruntime/infer_demo.py",
"diff": "@@ -38,8 +38,12 @@ class PicoDet():\nso = ort.SessionOptions()\nso.log_severity_level = 3\nself.net = ort.InferenceSession(model_pb_path, so)\n- self.input_shape = (self.net.get_inputs()[0].shape[2],\n- self.net.get_inputs()[0].shape[3])\n+ inputs_name = [a.name for a in self.net.get_inputs()]\n+ inputs_shape = {\n+ k: v.shape\n+ for k, v in zip(inputs_name, self.net.get_inputs())\n+ }\n+ self.input_shape = inputs_shape['image'][2:]\ndef _normalize(self, img):\nimg = img.astype(np.float32)\n@@ -51,6 +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+ ]).astype('float32')\nscale_factor = np.array([[im_scale_y, im_scale_x]]).astype('float32')\nif keep_ratio and srcimg.shape[0] != srcimg.shape[1]:\n@@ -87,7 +93,7 @@ class PicoDet():\nimg = cv2.resize(\nsrcimg, self.input_shape, interpolation=cv2.INTER_AREA)\n- return img, scale_factor\n+ return img, img_shape, scale_factor\ndef get_color_map_list(self, num_classes):\ncolor_map = num_classes * [0, 0, 0]\n@@ -104,15 +110,20 @@ class PicoDet():\nreturn color_map\ndef detect(self, srcimg):\n- img, scale_factor = self.resize_image(srcimg)\n+ img, im_shape, scale_factor = self.resize_image(srcimg)\nimg = self._normalize(img)\nblob = np.expand_dims(np.transpose(img, (2, 0, 1)), axis=0)\n- outs = self.net.run(None, {\n- self.net.get_inputs()[0].name: blob,\n- self.net.get_inputs()[1].name: scale_factor\n- })\n+ inputs_dict = {\n+ 'im_shape': im_shape,\n+ 'image': blob,\n+ 'scale_factor': scale_factor\n+ }\n+ inputs_name = [a.name for a in self.net.get_inputs()]\n+ net_inputs = {k: inputs_dict[k] for k in inputs_name}\n+\n+ outs = self.net.run(None, net_inputs)\nouts = np.array(outs[0])\nexpect_boxes = (outs[:, 1] > 0.5) & (outs[:, 0] > -1)\n@@ -181,7 +192,7 @@ if __name__ == '__main__':\nparser.add_argument(\n\"--img_fold\", dest=\"img_fold\", type=str, default=\"./imgs\")\nparser.add_argument(\n- \"--result_fold\", dest=\"result_fold\", type=str, default=\"./results\")\n+ \"--result_fold\", dest=\"result_fold\", type=str, default=\"results\")\nargs = parser.parse_args()\nnet = PicoDet(\n@@ -191,3 +202,6 @@ if __name__ == '__main__':\niou_threshold=args.nmsThreshold)\nnet.detect_folder(args.img_fold, args.result_fold)\n+ print(\n+ f'infer results in ./deploy/third_engine/demo_onnxruntime/{args.result_fold}'\n+ )\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/picodet/picodet_s_320_coco_lcnet_linux_gpu_normal_normal_paddle2onnx_python_linux_gpu_cpu.txt",
"diff": "+===========================paddle2onnx_params===========================\n+model_name:picodet_s_320_coco_lcnet\n+python:python3.7\n+filename:null\n+##\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+quant_export:tools/export_model.py -c configs/picodet/picodet_s_320_coco_lcnet.yml --slim_config configs/picodet/picodet_s_320_coco_lcnet.yml -o\n+fpgm_export:tools/export_model.py -c configs/picodet/picodet_s_320_coco_lcnet.yml --slim_config configs/picodet/picodet_s_320_coco_lcnet.yml -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 configs/picodet/picodet_s_320_coco_lcnet.yml -o\n+##\n+2onnx: paddle2onnx\n+--model_dir:./output_inference/picodet_s_320_coco_lcnet/\n+--model_filename:model.pdmodel\n+--params_filename:model.pdiparams\n+--save_file:./deploy/third_engine/demo_onnxruntime/onnx_file/picodet_s_320_coco.onnx\n+--opset_version:11\n+##\n+inference:infer_demo.py\n+--modelpath:./onnx_file/picodet_s_320_coco.onnx\n+--img_fold:./imgs\n+--result_fold:results\n+infer_mode:norm\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": "@@ -4,7 +4,7 @@ source test_tipc/utils_func.sh\nFILENAME=$1\n# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer'\n# 'whole_train_whole_infer', 'whole_infer', 'klquant_whole_infer',\n-# 'cpp_infer', 'serving_infer', 'lite_infer']\n+# 'cpp_infer', 'serving_infer', 'lite_infer', 'paddle2onnx_infer']\nMODE=$2\n# parse params\n@@ -67,6 +67,10 @@ elif [ ${MODE} = \"benchmark_train\" ];then\nwget -nc -P ./dataset/mot/ https://paddledet.bj.bcebos.com/data/mot_benchmark.tar\ncd ./dataset/mot/ && tar -xvf mot_benchmark.tar && mv -u mot_benchmark/* .\nrm -rf mot_benchmark/ && cd ../../\n+elif [ ${MODE} = \"paddle2onnx_infer\" ];then\n+ # set paddle2onnx_infer enve\n+ ${python} -m pip install install paddle2onnx\n+ ${python} -m pip install onnxruntime==1.10.0\nelse\n# download coco lite data\nwget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/tipc/coco_tipc.tar\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/test_paddle2onnx.sh",
"diff": "+#!/bin/bash\n+source test_tipc/utils_func.sh\n+\n+FILENAME=$1\n+\n+# parser model_name\n+dataline=$(cat ${FILENAME})\n+IFS=$'\\n'\n+lines=(${dataline})\n+model_name=$(func_parser_value \"${lines[1]}\")\n+echo \"ppdet onnx_infer: ${model_name}\"\n+python=$(func_parser_value \"${lines[2]}\")\n+filename_key=$(func_parser_key \"${lines[3]}\")\n+filename_value=$(func_parser_value \"${lines[3]}\")\n+\n+# export params\n+save_export_key=$(func_parser_key \"${lines[5]}\")\n+save_export_value=$(func_parser_value \"${lines[5]}\")\n+export_weight_key=$(func_parser_key \"${lines[6]}\")\n+export_weight_value=$(func_parser_value \"${lines[6]}\")\n+norm_export=$(func_parser_value \"${lines[7]}\")\n+pact_export=$(func_parser_value \"${lines[8]}\")\n+fpgm_export=$(func_parser_value \"${lines[9]}\")\n+distill_export=$(func_parser_value \"${lines[10]}\")\n+export_key1=$(func_parser_key \"${lines[11]}\")\n+export_value1=$(func_parser_value \"${lines[11]}\")\n+export_key2=$(func_parser_key \"${lines[12]}\")\n+export_value2=$(func_parser_value \"${lines[12]}\")\n+kl_quant_export=$(func_parser_value \"${lines[13]}\")\n+\n+# parser paddle2onnx\n+padlle2onnx_cmd=$(func_parser_value \"${lines[15]}\")\n+infer_model_dir_key=$(func_parser_key \"${lines[16]}\")\n+infer_model_dir_value=$(func_parser_value \"${lines[16]}\")\n+model_filename_key=$(func_parser_key \"${lines[17]}\")\n+model_filename_value=$(func_parser_value \"${lines[17]}\")\n+params_filename_key=$(func_parser_key \"${lines[18]}\")\n+params_filename_value=$(func_parser_value \"${lines[18]}\")\n+save_file_key=$(func_parser_key \"${lines[19]}\")\n+save_file_value=$(func_parser_value \"${lines[19]}\")\n+opset_version_key=$(func_parser_key \"${lines[20]}\")\n+opset_version_value=$(func_parser_value \"${lines[20]}\")\n+\n+# parser onnx inference\n+inference_py=$(func_parser_value \"${lines[22]}\")\n+model_file_key=$(func_parser_key \"${lines[23]}\")\n+model_file_value=$(func_parser_value \"${lines[23]}\")\n+img_fold_key=$(func_parser_key \"${lines[24]}\")\n+img_fold_value=$(func_parser_value \"${lines[24]}\")\n+results_fold_key=$(func_parser_key \"${lines[25]}\")\n+results_fold_value=$(func_parser_value \"${lines[25]}\")\n+onnx_infer_mode_list=$(func_parser_value \"${lines[26]}\")\n+\n+LOG_PATH=\"./test_tipc/output\"\n+mkdir -p ${LOG_PATH}\n+status_log=\"${LOG_PATH}/results_paddle2onnx.log\"\n+\n+function func_paddle2onnx(){\n+ IFS='|'\n+ _script=$1\n+\n+ # paddle2onnx\n+ echo \"################### run onnx export ###################\"\n+ _save_log_path=\"${LOG_PATH}/paddle2onnx_infer_cpu.log\"\n+ set_dirname=$(func_set_params \"${infer_model_dir_key}\" \"${infer_model_dir_value}\")\n+ set_model_filename=$(func_set_params \"${model_filename_key}\" \"${model_filename_value}\")\n+ set_params_filename=$(func_set_params \"${params_filename_key}\" \"${params_filename_value}\")\n+ set_save_model=$(func_set_params \"${save_file_key}\" \"${save_file_value}\")\n+ set_opset_version=$(func_set_params \"${opset_version_key}\" \"${opset_version_value}\")\n+ trans_model_cmd=\"${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version}\"\n+ eval $trans_model_cmd\n+ last_status=${PIPESTATUS[0]}\n+ status_check $last_status \"${trans_model_cmd}\" \"${status_log}\"\n+ # python inference\n+ echo \"################### run infer ###################\"\n+ cd ./deploy/third_engine/demo_onnxruntime/\n+ model_file=$(func_set_params \"${model_file_key}\" \"${model_file_value}\")\n+ img_fold=$(func_set_params \"${img_fold_key}\" \"${img_fold_value}\")\n+ results_fold=$(func_set_params \"${results_fold_key}\" \"${results_fold_value}\")\n+ infer_model_cmd=\"${python} ${inference_py} ${model_file} ${img_fold} ${results_fold}\"\n+ eval $infer_model_cmd\n+ last_status=${PIPESTATUS[0]}\n+ status_check $last_status \"${infer_model_cmd}\" \"${status_log}\"\n+}\n+\n+export Count=0\n+IFS=\"|\"\n+echo \"################### run paddle export ###################\"\n+for infer_mode in ${onnx_infer_mode_list[*]}; do\n+\n+ # run export\n+ case ${infer_mode} in\n+ norm) run_export=${norm_export} ;;\n+ quant) run_export=${pact_export} ;;\n+ fpgm) run_export=${fpgm_export} ;;\n+ distill) run_export=${distill_export} ;;\n+ kl_quant) run_export=${kl_quant_export} ;;\n+ *) echo \"Undefined infer_mode!\"; exit 1;\n+ esac\n+ if [ ${run_export} = \"null\" ]; then\n+ continue\n+ fi\n+ set_export_weight=$(func_set_params \"${export_weight_key}\" \"${export_weight_value}\")\n+ set_save_export_dir=$(func_set_params \"${save_export_key}\" \"${save_export_value}\")\n+ set_filename=$(func_set_params \"${filename_key}\" \"${model_name}\")\n+ export_cmd=\"${python} ${run_export} ${set_export_weight} ${set_filename} ${set_save_export_dir} \"\n+ echo $export_cmd\n+ eval $export_cmd\n+ status_export=$?\n+ status_check $status_export \"${export_cmd}\" \"${status_log}\"\n+done\n+func_paddle2onnx\n\\ No newline at end of file\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[test_tipc] add paddle2onnx; fix onnxruntime infer_demo (#5857)
Co-authored-by: lyuwenyu <wenyu.lyu@gmail.com>
|
499,333 |
29.04.2022 13:50:23
| -28,800 |
e22f2d6b5ab83a6baaae8ced11dfdefe0b01795e
|
fix mask rcnn target
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/proposal_generator/target.py",
"new_path": "ppdet/modeling/proposal_generator/target.py",
"diff": "@@ -356,7 +356,7 @@ def generate_mask_target(gt_segms, rois, labels_int32, sampled_gt_inds,\nfg_inds_new = fg_inds.reshape([-1]).numpy()\nresults = []\nif len(gt_segms_per_im) > 0:\n- for j in fg_inds_new:\n+ for j in range(fg_inds_new.shape[0]):\nresults.append(\nrasterize_polygons_within_box(new_segm[j], boxes[j],\nresolution))\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix mask rcnn target (#5861)
|
499,299 |
29.04.2022 14:51:30
| -28,800 |
bb319a8a598cdaa3fdf431988f28cafbdd9b2a11
|
add pdparams links
|
[
{
"change_type": "MODIFY",
"old_path": "configs/keypoint/README_en.md",
"new_path": "configs/keypoint/README_en.md",
"diff": "@@ -89,6 +89,13 @@ MPII Dataset\n| :---- | -------- | :--------: | :------------: | :----------------------------------------------------------: | -------------------------------------------- |\n| HRNet-w32 | 256x256 | 90.6 | 38.5 | [hrnet_w32_256x256_mpii.pdparams](https://paddledet.bj.bcebos.com/models/keypoint/hrnet_w32_256x256_mpii.pdparams) | [config](./hrnet/hrnet_w32_256x256_mpii.yml) |\n+\n+Model for Scenes\n+| Model | Strategy | Input Size | Precision | Inference Speed |Model Weights | Model Inference and Deployment | description|\n+| :---- | ---|----- | :--------: | :-------: |:------------: |:------------: |:-------------------: |\n+| HRNet-w32 + DarkPose | Top-Down|256x192 | AP: 87.1 (on internal dataset)| 2.9ms per person |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.pdparams) |[Link](https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.zip) | Especially optimized for fall scenarios, the model is applied to [PP-Human](../../deploy/pphuman/README_en.md) |\n+\n+\nWe also release [PP-TinyPose](./tiny_pose/README_en.md), a real-time keypoint detection model optimized for mobile devices. Welcome to experience.\n## Getting Start\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add pdparams links (#5864)
|
499,299 |
29.04.2022 18:27:39
| -28,800 |
55ae0d1b83f937c316c6788d834b9a35f1cd99ae
|
fix kpt doc error
|
[
{
"change_type": "MODIFY",
"old_path": "configs/keypoint/README_en.md",
"new_path": "configs/keypoint/README_en.md",
"diff": "@@ -161,7 +161,6 @@ CUDA_VISIBLE_DEVICES=0 python3 tools/infer.py -c configs/keypoint/higherhrnet/hi\n```shell\n#Export Detection Model\n-\npython tools/export_model.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/ppyolov2_r50vd_dcn_365e_coco.pdparams\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix kpt doc error (#5867)
|
499,333 |
01.05.2022 15:17:21
| -28,800 |
3d1e2320bf5b6ae30eb5c0fd296e83b8bd88afb4
|
[test_tipc] add serving tipc
|
[
{
"change_type": "ADD",
"old_path": null,
"new_path": "test_tipc/configs/yolov3/yolov3_darknet53_270e_coco_model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt",
"diff": "+===========================cpp_infer_params===========================\n+model_name:yolov3_darknet53_270e_coco\n+python:python\n+filename:null\n+##\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+quant_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+--export_serving_model:True\n+##\n+start_serving:-m paddle_serving_server.serve --model serving_server\n+--port:9393\n+--gpu_ids:0\n+##\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/prepare.sh",
"new_path": "test_tipc/prepare.sh",
"diff": "@@ -71,6 +71,16 @@ elif [ ${MODE} = \"paddle2onnx_infer\" ];then\n# set paddle2onnx_infer enve\n${python} -m pip install install paddle2onnx\n${python} -m pip install onnxruntime==1.10.0\n+elif [ ${MODE} = \"serving_infer\" ];then\n+ git clone https://github.com/PaddlePaddle/Serving\n+ bash Serving/tools/paddle_env_install.sh\n+ cd Serving\n+ pip install -r python/requirements.txt\n+ cd ..\n+ pip install paddle-serving-client==0.8.3 -i https://pypi.tuna.tsinghua.edu.cn/simple\n+ pip install paddle-serving-app==0.8.3 -i https://pypi.tuna.tsinghua.edu.cn/simple\n+ pip install paddle-serving-server-gpu==0.8.3.post101 -i https://pypi.tuna.tsinghua.edu.cn/simple\n+ python -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\nwget -nc -P ./dataset/coco/ https://paddledet.bj.bcebos.com/data/tipc/coco_tipc.tar\n"
},
{
"change_type": "MODIFY",
"old_path": "test_tipc/test_serving.sh",
"new_path": "test_tipc/test_serving.sh",
"diff": "+#!/bin/bash\n+source test_tipc/utils_func.sh\n+\n+FILENAME=$1\n+\n+# parser model_name\n+dataline=$(cat ${FILENAME})\n+IFS=$'\\n'\n+lines=(${dataline})\n+model_name=$(func_parser_value \"${lines[1]}\")\n+echo \"ppdet serving: ${model_name}\"\n+python=$(func_parser_value \"${lines[2]}\")\n+filename_key=$(func_parser_key \"${lines[3]}\")\n+filename_value=$(func_parser_value \"${lines[3]}\")\n+\n+# export params\n+save_export_key=$(func_parser_key \"${lines[5]}\")\n+save_export_value=$(func_parser_value \"${lines[5]}\")\n+export_weight_key=$(func_parser_key \"${lines[6]}\")\n+export_weight_value=$(func_parser_value \"${lines[6]}\")\n+norm_export=$(func_parser_value \"${lines[7]}\")\n+pact_export=$(func_parser_value \"${lines[8]}\")\n+fpgm_export=$(func_parser_value \"${lines[9]}\")\n+distill_export=$(func_parser_value \"${lines[10]}\")\n+export_key1=$(func_parser_key \"${lines[11]}\")\n+export_value1=$(func_parser_value \"${lines[11]}\")\n+export_key2=$(func_parser_key \"${lines[12]}\")\n+export_value2=$(func_parser_value \"${lines[12]}\")\n+kl_quant_export=$(func_parser_value \"${lines[13]}\")\n+export_serving_model_key=$(func_parser_key \"${lines[14]}\")\n+export_serving_model_value=$(func_parser_value \"${lines[14]}\")\n+# parser serving\n+start_serving=$(func_parser_value \"${lines[16]}\")\n+port_key=$(func_parser_key \"${lines[17]}\")\n+port_value=$(func_parser_value \"${lines[17]}\")\n+gpu_id_key=$(func_parser_key \"${lines[18]}\")\n+gpu_id_value=$(func_parser_value \"${lines[18]}\")\n+\n+LOG_PATH=\"./test_tipc/output\"\n+mkdir -p ${LOG_PATH}\n+status_log=\"${LOG_PATH}/results_serving.log\"\n+\n+function func_serving(){\n+ IFS='|'\n+ if [ ${gpu_id_key} = \"null\" ]; then\n+ start_serving_command=\"nohup ${python} ${start_serving} ${port_key} ${port_value} > serving.log 2>&1 &\"\n+ else\n+ start_serving_command=\"nohup ${python} ${start_serving} ${port_key} ${port_value} ${gpu_id_key} ${gpu_id_value} > serving.log 2>&1 &\"\n+ fi\n+ echo $start_serving_command\n+ eval $start_serving_command\n+ last_status=${PIPESTATUS[0]}\n+ status_check $last_status \"${start_serving_command}\" \"${status_log}\"\n+}\n+cd output_inference/${model_name}\n+echo $PWD\n+func_serving\n+test_command=\"${python} ../../deploy/serving/test_client.py ../../deploy/serving/label_list.txt ../../demo/000000014439.jpg\"\n+echo $test_command\n+eval $test_command\n+last_status=${PIPESTATUS[0]}\n+status_check $last_status\"${test_command}\" \"${status_log}\"\n+\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
[test_tipc] add serving tipc (#5865)
|
499,298 |
04.05.2022 23:25:00
| -28,800 |
2b88308f9b97bba1443688e6a6907cb02cb50fb8
|
fix pphuman vis when nothing detected
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/pphuman/pipeline.py",
"new_path": "deploy/pphuman/pipeline.py",
"diff": "@@ -537,8 +537,9 @@ class PipePredictor(object):\nself.pipe_timer.total_time.end()\nif self.cfg['visual']:\n_, _, fps = self.pipe_timer.get_total_time()\n- im = self.visualize_video(frame, mot_res, frame_id,\n- fps) # visualize\n+ im = self.visualize_video(frame, mot_res, frame_id, fps,\n+ entrance, records,\n+ center_traj) # visualize\nwriter.write(im)\nif self.file_name is None: # use camera_id\ncv2.imshow('PPHuman', im)\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix pphuman vis when nothing detected (#5858)
|
499,299 |
07.05.2022 10:06:57
| -28,800 |
756e3dca26e5c4f853203fcc8f5b5af1da234eeb
|
fix cpp infer error in blazeface
|
[
{
"change_type": "MODIFY",
"old_path": "deploy/cpp/src/object_detector.cc",
"new_path": "deploy/cpp/src/object_detector.cc",
"diff": "@@ -213,10 +213,6 @@ void ObjectDetector::Postprocess(\ncv::Mat raw_mat = mats[im_id];\nint rh = 1;\nint rw = 1;\n- if (config_.arch_ == \"Face\") {\n- rh = raw_mat.rows;\n- rw = raw_mat.cols;\n- }\nfor (int j = start_idx; j < start_idx + bbox_num[im_id]; j++) {\nif (is_rbox) {\n// Class id\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix cpp infer error in blazeface (#5892)
|
499,304 |
07.05.2022 13:37:59
| -28,800 |
acaf10684d8e071e48098c660355d54981d93b7c
|
add pp-yoloe quantization model
|
[
{
"change_type": "MODIFY",
"old_path": "configs/slim/README_en.md",
"new_path": "configs/slim/README_en.md",
"diff": "@@ -124,6 +124,8 @@ Description:\n| Model | Compression Strategy | Input Size | Model Volume(MB) | Prediction Delay(V100) | Prediction Delay(SD855) | Box AP | Download | Download of Inference Model | Model Configuration File | Compression Algorithm Configuration File |\n| ------------------------- | -------------------------- | ----------- | :--------------: | :--------------------: | :---------------------: | :-------------------: | :-----------------------------------------------------------------------------------------: | :-----------------------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------: |\n+| PP-YOLOE-l | baseline | 640 | - | 11.2ms(trt_fp32) | 7.7ms(trt_fp16) | -- | 50.9 | [link](https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams) | - | [Configuration File](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml) | - |\n+| PP-YOLOE-l | Common Online quantitative | 640 | - | 6.7ms(trt_int8) | -- | 48.8 | [link](https://paddledet.bj.bcebos.com/models/slim/ppyoloe_l_coco_qat.pdparams) | - | [Configuration File](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyoloe/ppyoloe_crn_l_300e_coco.yml) | [Configuration File](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/slim/quant/ppyoloe_l_qat.yml) |\n| PP-YOLOv2_R50vd | baseline | 640 | 208.6 | 19.1ms | -- | 49.1 | [link](https://paddledet.bj.bcebos.com/models/ppyolov2_r50vd_dcn_365e_coco.pdparams) | [link](https://paddledet.bj.bcebos.com/models/slim/ppyolov2_r50vd_dcn_365e_coco.tar) | [Configuration File](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | - |\n| PP-YOLOv2_R50vd | PACT Online quantitative | 640 | -- | 17.3ms | -- | 48.1 | [link](https://paddledet.bj.bcebos.com/models/slim/ppyolov2_r50vd_dcn_qat.pdparams) | [link](https://paddledet.bj.bcebos.com/models/slim/ppyolov2_r50vd_dcn_qat.tar) | [Configuration File ](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | [Configuration File ](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/slim/quant/ppyolov2_r50vd_dcn_qat.yml) |\n| PP-YOLO_R50vd | baseline | 608 | 183.3 | 17.4ms | -- | 44.8 | [link](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [link](https://paddledet.bj.bcebos.com/models/slim/ppyolo_r50vd_dcn_1x_coco.tar) | [Configuration File ](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | - |\n"
},
{
"change_type": "ADD",
"old_path": null,
"new_path": "configs/slim/quant/ppyoloe_l_qat.yml",
"diff": "+pretrain_weights: https://paddledet.bj.bcebos.com/models/ppyoloe_crn_l_300e_coco.pdparams\n+slim: QAT\n+\n+QAT:\n+ quant_config: {\n+ 'weight_quantize_type': 'channel_wise_abs_max', 'activation_quantize_type': 'moving_average_abs_max',\n+ 'weight_bits': 8, 'activation_bits': 8, 'dtype': 'int8', 'window_size': 10000, 'moving_rate': 0.9,\n+ 'quantizable_layer_type': ['Conv2D', 'Linear']}\n+ print_model: True\n+\n+epoch: 30\n+snapshot_epoch: 5\n+LearningRate:\n+ base_lr: 0.001\n+ schedulers:\n+ - !PiecewiseDecay\n+ gamma: 0.1\n+ milestones:\n+ - 10\n+ - 20\n+ - !LinearWarmup\n+ start_factor: 0.\n+ steps: 100\n+\n+TrainReader:\n+ batch_size: 8\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/engine/trainer.py",
"new_path": "ppdet/engine/trainer.py",
"diff": "@@ -701,6 +701,7 @@ class Trainer(object):\nif hasattr(self.model, 'deploy'):\nself.model.deploy = True\n+ if 'slim' not in self.cfg:\nfor layer in self.model.sublayers():\nif hasattr(layer, 'convert_to_deploy'):\nlayer.convert_to_deploy()\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/ppyoloe_head.py",
"new_path": "ppdet/modeling/heads/ppyoloe_head.py",
"diff": "@@ -111,6 +111,7 @@ class PPYOLOEHead(nn.Layer):\nin_c, 4 * (self.reg_max + 1), 3, padding=1))\n# projection conv\nself.proj_conv = nn.Conv2D(self.reg_max + 1, 1, 1, bias_attr=False)\n+ self.proj_conv.skip_quant = True\nself._init_weights()\n@classmethod\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/slim/quant.py",
"new_path": "ppdet/slim/quant.py",
"diff": "@@ -38,6 +38,11 @@ class QAT(object):\nlogger.info(\"Model before quant:\")\nlogger.info(model)\n+ # For PP-YOLOE, convert model to deploy firstly.\n+ for layer in model.sublayers():\n+ if hasattr(layer, 'convert_to_deploy'):\n+ layer.convert_to_deploy()\n+\nself.quanter.quantize(model)\nif self.print_model:\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add pp-yoloe quantization model (#5894)
|
499,304 |
10.05.2022 09:44:03
| -28,800 |
2ecf726e6a44fbbf16094f6b8c8a76410d0f29e8
|
fix picofeat assert error
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/heads/pico_head.py",
"new_path": "ppdet/modeling/heads/pico_head.py",
"diff": "@@ -91,7 +91,7 @@ class PicoFeat(nn.Layer):\nself.reg_convs = []\nif use_se:\nassert share_cls_reg == True, \\\n- 'In the case of using se, share_cls_reg is not supported'\n+ 'In the case of using se, share_cls_reg must be set to True'\nself.se = nn.LayerList()\nfor stage_idx in range(num_fpn_stride):\ncls_subnet_convs = []\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
fix picofeat assert error (#5911)
|
499,301 |
10.05.2022 17:23:06
| -28,800 |
81f84e237b71fb152d749a06860d34d34938dde1
|
add slim link
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/README.md",
"new_path": "configs/ppyoloe/README.md",
"diff": "@@ -136,6 +136,7 @@ PP-YOLOE can be deployed by following approches:\n- Paddle Inference [Python](../../deploy/python) & [C++](../../deploy/cpp)\n- [Paddle-TensorRT](../../deploy/TENSOR_RT.md)\n- [PaddleServing](https://github.com/PaddlePaddle/Serving)\n+ - [PaddleSlim](../configs/slim)\nNext, we will introduce how to use Paddle Inference to deploy PP-YOLOE models in TensorRT FP16 mode.\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
add slim link (#5908)
|
499,333 |
10.05.2022 19:12:33
| -28,800 |
a853016850a3a0a173544d7072bd8cc115dd279e
|
refine device when assign on cpu
|
[
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/post_process.py",
"new_path": "ppdet/modeling/post_process.py",
"diff": "@@ -194,8 +194,7 @@ class MaskPostProcess(object):\nsuper(MaskPostProcess, self).__init__()\nself.binary_thresh = binary_thresh\nself.export_onnx = export_onnx\n- self.assign_on_cpu = assign_on_cpu and paddle.device.is_compiled_with_cuda(\n- )\n+ self.assign_on_cpu = assign_on_cpu\ndef paste_mask(self, masks, boxes, im_h, im_w):\n\"\"\"\n@@ -240,6 +239,7 @@ class MaskPostProcess(object):\n\"\"\"\nnum_mask = mask_out.shape[0]\norigin_shape = paddle.cast(origin_shape, 'int32')\n+ device = paddle.device.get_device()\nif self.export_onnx:\nh, w = origin_shape[0][0], origin_shape[0][1]\n@@ -269,7 +269,7 @@ class MaskPostProcess(object):\nim_w] = pred_mask\nid_start += bbox_num[i]\nif self.assign_on_cpu:\n- paddle.set_device('gpu')\n+ paddle.set_device(device)\nreturn pred_result\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/proposal_generator/target.py",
"new_path": "ppdet/modeling/proposal_generator/target.py",
"diff": "@@ -74,9 +74,11 @@ def label_box(anchors,\nis_crowd=None,\nassign_on_cpu=False):\nif assign_on_cpu:\n+ device = paddle.device.get_device()\npaddle.set_device(\"cpu\")\niou = bbox_overlaps(gt_boxes, anchors)\n- paddle.set_device(\"gpu\")\n+ paddle.set_device(device)\n+\nelse:\niou = bbox_overlaps(gt_boxes, anchors)\nn_gt = gt_boxes.shape[0]\n"
},
{
"change_type": "MODIFY",
"old_path": "ppdet/modeling/proposal_generator/target_layer.py",
"new_path": "ppdet/modeling/proposal_generator/target_layer.py",
"diff": "@@ -68,8 +68,7 @@ class RPNTargetAssign(object):\nself.negative_overlap = negative_overlap\nself.ignore_thresh = ignore_thresh\nself.use_random = use_random\n- self.assign_on_cpu = assign_on_cpu and paddle.device.is_compiled_with_cuda(\n- )\n+ self.assign_on_cpu = assign_on_cpu\ndef __call__(self, inputs, anchors):\n\"\"\"\n@@ -150,8 +149,7 @@ class BBoxAssigner(object):\nself.use_random = use_random\nself.cascade_iou = cascade_iou\nself.num_classes = num_classes\n- self.assign_on_cpu = assign_on_cpu and paddle.device.is_compiled_with_cuda(\n- )\n+ self.assign_on_cpu = assign_on_cpu\ndef __call__(self,\nrpn_rois,\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
refine device when assign on cpu (#5924)
|
499,301 |
12.05.2022 10:13:54
| -28,800 |
2629454d0c87fc0d2295b11ffa529829e5fb4c5f
|
update e quant link
|
[
{
"change_type": "MODIFY",
"old_path": "configs/ppyoloe/README.md",
"new_path": "configs/ppyoloe/README.md",
"diff": "@@ -136,7 +136,7 @@ PP-YOLOE can be deployed by following approches:\n- Paddle Inference [Python](../../deploy/python) & [C++](../../deploy/cpp)\n- [Paddle-TensorRT](../../deploy/TENSOR_RT.md)\n- [PaddleServing](https://github.com/PaddlePaddle/Serving)\n- - [PaddleSlim](../configs/slim)\n+ - [PaddleSlim](../slim)\nNext, we will introduce how to use Paddle Inference to deploy PP-YOLOE models in TensorRT FP16 mode.\n"
}
] |
Python
|
Apache License 2.0
|
paddlepaddle/paddledetection
|
update e quant link (#5938)
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.