File size: 6,874 Bytes
0a82b18
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
import os
import torch
import py3_wget
from pathlib import Path
import torchvision.transforms as tfm
from kornia.feature import DeDoDe
import kornia

from matching import get_version, THIRD_PARTY_DIR, WEIGHTS_DIR, BaseMatcher
from matching.utils import add_to_path, resize_to_divisible

add_to_path(THIRD_PARTY_DIR.joinpath("DeDoDe"))

from DeDoDe import dedode_detector_L, dedode_descriptor_G
from DeDoDe.matchers.dual_softmax_matcher import DualSoftMaxMatcher


class DedodeMatcher(BaseMatcher):
    detector_path = WEIGHTS_DIR.joinpath("dedode_detector_L.pth")
    detector_v2_path = WEIGHTS_DIR.joinpath("dedode_detector_L_v2.pth")
    descriptor_path = WEIGHTS_DIR.joinpath("dedode_descriptor_G.pth")
    dino_patch_size = 14

    def __init__(self, device="cpu", max_num_keypoints=2048, dedode_thresh=0.05, detector_version=2, *args, **kwargs):
        super().__init__(device, **kwargs)

        if self.device != "cuda": # only cuda devices work due to autocast in cuda in upstream.
            raise ValueError("Only device 'cuda' supported for DeDoDe.")

        self.max_keypoints = max_num_keypoints
        self.threshold = dedode_thresh
        self.normalize = tfm.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])

        self.download_weights()

        detector_weight_path = self.detector_path if detector_version == 1 else self.detector_v2_path
        self.detector = dedode_detector_L(weights=torch.load(detector_weight_path, map_location=device), device=device)
        self.descriptor = dedode_descriptor_G(
            weights=torch.load(self.descriptor_path, map_location=device), device=device
        )
        self.matcher = DualSoftMaxMatcher()

    @staticmethod
    def download_weights():
        detector_url = (
            "https://github.com/Parskatt/DeDoDe/releases/download/dedode_pretrained_models/dedode_detector_L.pth"
        )
        detector_v2_url = "https://github.com/Parskatt/DeDoDe/releases/download/v2/dedode_detector_L_v2.pth"
        descr_url = (
            "https://github.com/Parskatt/DeDoDe/releases/download/dedode_pretrained_models/dedode_descriptor_G.pth"
        )
        os.makedirs("model_weights", exist_ok=True)
        if not os.path.isfile(DedodeMatcher.detector_path):
            print("Downloading dedode_detector_L.pth")
            py3_wget.download_file(detector_url, DedodeMatcher.detector_path)

        if not os.path.isfile(DedodeMatcher.detector_v2_path):
            print("Downloading dedode_descriptor_L-v2.pth")
            py3_wget.download_file(detector_v2_url, DedodeMatcher.detector_v2_path)

        if not os.path.isfile(DedodeMatcher.descriptor_path):
            print("Downloading dedode_descriptor_G.pth")
            py3_wget.download_file(descr_url, DedodeMatcher.descriptor_path)

    def preprocess(self, img):
        # ensure that the img has the proper w/h to be compatible with patch sizes
        _, h, w = img.shape
        orig_shape = h, w
        img = resize_to_divisible(img, self.dino_patch_size)

        img = self.normalize(img).unsqueeze(0).to(self.device)
        return img, orig_shape

    def _forward(self, img0, img1):
        img0, img0_orig_shape = self.preprocess(img0)
        img1, img1_orig_shape = self.preprocess(img1)

        batch_0 = {"image": img0}
        detections_0 = self.detector.detect(batch_0, num_keypoints=self.max_keypoints)
        keypoints_0, P_0 = detections_0["keypoints"], detections_0["confidence"]

        batch_1 = {"image": img1}
        detections_1 = self.detector.detect(batch_1, num_keypoints=self.max_keypoints)
        keypoints_1, P_1 = detections_1["keypoints"], detections_1["confidence"]

        description_0 = self.descriptor.describe_keypoints(batch_0, keypoints_0)["descriptions"]
        description_1 = self.descriptor.describe_keypoints(batch_1, keypoints_1)["descriptions"]

        matches_0, matches_1, _ = self.matcher.match(
            keypoints_0,
            description_0,
            keypoints_1,
            description_1,
            P_A=P_0,
            P_B=P_1,
            normalize=True,
            inv_temp=20,
            threshold=self.threshold,  # Increasing threshold -> fewer matches, fewer outliers
        )

        H0, W0, H1, W1 = *img0.shape[-2:], *img1.shape[-2:]
        mkpts0, mkpts1 = self.matcher.to_pixel_coords(matches_0, matches_1, H0, W0, H1, W1)

        keypoints_0, keypoints_1 = self.matcher.to_pixel_coords(
            keypoints_0.squeeze(0), keypoints_1.squeeze(0), H0, W0, H1, W1
        )

        # dedode sometimes requires reshaping an image to fit vit patch size evenly, so we need to
        # rescale kpts to the original img
        keypoints_0 = self.rescale_coords(keypoints_0, *img0_orig_shape, H0, W0)
        keypoints_1 = self.rescale_coords(keypoints_1, *img1_orig_shape, H1, W1)

        mkpts0 = self.rescale_coords(mkpts0, *img0_orig_shape, H0, W0)
        mkpts1 = self.rescale_coords(mkpts1, *img1_orig_shape, H1, W1)

        return mkpts0, mkpts1, keypoints_0, keypoints_1, description_0.squeeze(0), description_1.squeeze(0)


class DedodeKorniaMatcher(BaseMatcher):
    def __init__(
        self,
        device="cpu",
        max_num_keypoints=2048,
        detector_weights="L-C4-v2",
        descriptor_weights="G-C4",
        match_thresh=0.05,
        *args,
        **kwargs,
    ):
        super().__init__(device, **kwargs)

        major, minor, patch = get_version(kornia)
        assert major > 1 or (
            minor >= 7 and patch >= 3
        ), "DeDoDeKornia only available in kornia v 0.7.3 or greater. Update kornia to use this model."

        self.max_keypoints = max_num_keypoints

        self.model = DeDoDe.from_pretrained(
            detector_weights=detector_weights,
            descriptor_weights=descriptor_weights,
            amp_dtype=torch.float32 if device != "cuda" else torch.float16,
        )
        self.model.to(device)
        self.matcher = DualSoftMaxMatcher()

        self.threshold = match_thresh

    def preprocess(self, img):
        if img.ndim == 3:
            return img[None]
        else:
            return img

    @torch.inference_mode()
    def _forward(self, img0, img1):
        img0 = self.preprocess(img0)
        img1 = self.preprocess(img1)

        keypoints_0, P_0, description_0 = self.model(img0, n=self.max_keypoints)
        keypoints_1, P_1, description_1 = self.model(img1, n=self.max_keypoints)

        mkpts0, mkpts1, _ = self.matcher.match(
            keypoints_0,
            description_0,
            keypoints_1,
            description_1,
            P_A=P_0,
            P_B=P_1,
            normalize=True,
            inv_temp=20,
            threshold=self.threshold,  # Increasing threshold -> fewer matches, fewer outliers
        )

        return mkpts0, mkpts1, keypoints_0, keypoints_1, description_0, description_1