File size: 10,096 Bytes
40c4807
 
8320ccc
 
 
 
 
 
 
 
40c4807
8320ccc
40c4807
 
8320ccc
 
 
 
40c4807
8320ccc
40c4807
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de8bee7
40c4807
 
5e02733
 
 
 
 
 
 
 
 
40c4807
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de8bee7
 
 
 
 
8320ccc
40c4807
 
 
 
 
de8bee7
 
40c4807
 
 
 
5e02733
40c4807
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8320ccc
 
 
40c4807
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8320ccc
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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
import warnings
from pathlib import Path
from typing import Any, Dict, Optional

import cv2
import matplotlib.pyplot as plt
import numpy as np
import torch

from hloc import extract_features, logger, match_dense, match_features
from hloc.utils.viz import add_text, plot_keypoints

from .utils import (
    ROOT,
    filter_matches,
    get_feature_model,
    get_model,
    load_config,
)
from .viz import display_matches, fig2im, plot_images

warnings.simplefilter("ignore")


class ImageMatchingAPI(torch.nn.Module):
    default_conf = {
        "ransac": {
            "enable": True,
            "estimator": "poselib",
            "geometry": "homography",
            "method": "RANSAC",
            "reproj_threshold": 3,
            "confidence": 0.9999,
            "max_iter": 10000,
        },
    }

    def __init__(
        self,
        conf: dict = {},
        device: str = "cpu",
        detect_threshold: float = 0.015,
        max_keypoints: int = 1024,
        match_threshold: float = 0.2,
    ) -> None:
        """
        Initializes an instance of the ImageMatchingAPI class.

        Args:
            conf (dict): A dictionary containing the configuration parameters.
            device (str, optional): The device to use for computation. Defaults to "cpu".
            detect_threshold (float, optional): The threshold for detecting keypoints. Defaults to 0.015.
            max_keypoints (int, optional): The maximum number of keypoints to extract. Defaults to 1024.
            match_threshold (float, optional): The threshold for matching keypoints. Defaults to 0.2.

        Returns:
            None
        """
        super().__init__()
        self.device = device
        self.conf = {**self.default_conf, **conf}
        self._updata_config(detect_threshold, max_keypoints, match_threshold)
        self._init_models()
        if device == "cuda":
            memory_allocated = torch.cuda.memory_allocated(device)
            memory_reserved = torch.cuda.memory_reserved(device)
            logger.info(
                f"GPU memory allocated: {memory_allocated / 1024**2:.3f} MB"
            )
            logger.info(
                f"GPU memory reserved: {memory_reserved / 1024**2:.3f} MB"
            )
        self.pred = None

    def parse_match_config(self, conf):
        if conf["dense"]:
            return {
                **conf,
                "matcher": match_dense.confs.get(
                    conf["matcher"]["model"]["name"]
                ),
                "dense": True,
            }
        else:
            return {
                **conf,
                "feature": extract_features.confs.get(
                    conf["feature"]["model"]["name"]
                ),
                "matcher": match_features.confs.get(
                    conf["matcher"]["model"]["name"]
                ),
                "dense": False,
            }

    def _updata_config(
        self,
        detect_threshold: float = 0.015,
        max_keypoints: int = 1024,
        match_threshold: float = 0.2,
    ):
        self.dense = self.conf["dense"]
        if self.conf["dense"]:
            try:
                self.conf["matcher"]["model"][
                    "match_threshold"
                ] = match_threshold
            except TypeError as e:
                logger.error(e)
        else:
            self.conf["feature"]["model"]["max_keypoints"] = max_keypoints
            self.conf["feature"]["model"][
                "keypoint_threshold"
            ] = detect_threshold
            self.extract_conf = self.conf["feature"]

        self.match_conf = self.conf["matcher"]

    def _init_models(self):
        # initialize matcher
        self.matcher = get_model(self.match_conf)
        # initialize extractor
        if self.dense:
            self.extractor = None
        else:
            self.extractor = get_feature_model(self.conf["feature"])

    def _forward(self, img0, img1):
        if self.dense:
            pred = match_dense.match_images(
                self.matcher,
                img0,
                img1,
                self.match_conf["preprocessing"],
                device=self.device,
            )
            last_fixed = "{}".format(  # noqa: F841
                self.match_conf["model"]["name"]
            )
        else:
            pred0 = extract_features.extract(
                self.extractor, img0, self.extract_conf["preprocessing"]
            )
            pred1 = extract_features.extract(
                self.extractor, img1, self.extract_conf["preprocessing"]
            )
            pred = match_features.match_images(self.matcher, pred0, pred1)
        return pred

    @torch.inference_mode()
    def forward(
        self,
        img0: np.ndarray,
        img1: np.ndarray,
    ) -> Dict[str, np.ndarray]:
        """
        Forward pass of the image matching API.

        Args:
            img0: A 3D NumPy array of shape (H, W, C) representing the first image.
                  Values are in the range [0, 1] and are in RGB mode.
            img1: A 3D NumPy array of shape (H, W, C) representing the second image.
                  Values are in the range [0, 1] and are in RGB mode.

        Returns:
            A dictionary containing the following keys:
            - image0_orig: The original image 0.
            - image1_orig: The original image 1.
            - keypoints0_orig: The keypoints detected in image 0.
            - keypoints1_orig: The keypoints detected in image 1.
            - mkeypoints0_orig: The raw matches between image 0 and image 1.
            - mkeypoints1_orig: The raw matches between image 1 and image 0.
            - mmkeypoints0_orig: The RANSAC inliers in image 0.
            - mmkeypoints1_orig: The RANSAC inliers in image 1.
            - mconf: The confidence scores for the raw matches.
            - mmconf: The confidence scores for the RANSAC inliers.
        """
        # Take as input a pair of images (not a batch)
        assert isinstance(img0, np.ndarray)
        assert isinstance(img1, np.ndarray)
        self.pred = self._forward(img0, img1)
        if self.conf["ransac"]["enable"]:
            self.pred = self._geometry_check(self.pred)
        return self.pred

    def _geometry_check(
        self,
        pred: Dict[str, Any],
    ) -> Dict[str, Any]:
        """
        Filter matches using RANSAC. If keypoints are available, filter by keypoints.
        If lines are available, filter by lines. If both keypoints and lines are
        available, filter by keypoints.

        Args:
            pred (Dict[str, Any]): dict of matches, including original keypoints.
                                  See :func:`filter_matches` for the expected keys.

        Returns:
            Dict[str, Any]: filtered matches
        """
        pred = filter_matches(
            pred,
            ransac_method=self.conf["ransac"]["method"],
            ransac_reproj_threshold=self.conf["ransac"]["reproj_threshold"],
            ransac_confidence=self.conf["ransac"]["confidence"],
            ransac_max_iter=self.conf["ransac"]["max_iter"],
        )
        return pred

    def visualize(
        self,
        log_path: Optional[Path] = None,
    ) -> None:
        """
        Visualize the matches.

        Args:
            log_path (Path, optional): The directory to save the images. Defaults to None.

        Returns:
            None
        """
        if self.conf["dense"]:
            postfix = str(self.conf["matcher"]["model"]["name"])
        else:
            postfix = "{}_{}".format(
                str(self.conf["feature"]["model"]["name"]),
                str(self.conf["matcher"]["model"]["name"]),
            )
        titles = [
            "Image 0 - Keypoints",
            "Image 1 - Keypoints",
        ]
        pred: Dict[str, Any] = self.pred
        image0: np.ndarray = pred["image0_orig"]
        image1: np.ndarray = pred["image1_orig"]
        output_keypoints: np.ndarray = plot_images(
            [image0, image1], titles=titles, dpi=300
        )
        if (
            "keypoints0_orig" in pred.keys()
            and "keypoints1_orig" in pred.keys()
        ):
            plot_keypoints([pred["keypoints0_orig"], pred["keypoints1_orig"]])
            text: str = (
                f"# keypoints0: {len(pred['keypoints0_orig'])} \n"
                + f"# keypoints1: {len(pred['keypoints1_orig'])}"
            )
            add_text(0, text, fs=15)
        output_keypoints = fig2im(output_keypoints)
        # plot images with raw matches
        titles = [
            "Image 0 - Raw matched keypoints",
            "Image 1 - Raw matched keypoints",
        ]
        output_matches_raw, num_matches_raw = display_matches(
            pred, titles=titles, tag="KPTS_RAW"
        )
        # plot images with ransac matches
        titles = [
            "Image 0 - Ransac matched keypoints",
            "Image 1 - Ransac matched keypoints",
        ]
        output_matches_ransac, num_matches_ransac = display_matches(
            pred, titles=titles, tag="KPTS_RANSAC"
        )
        if log_path is not None:
            img_keypoints_path: Path = log_path / f"img_keypoints_{postfix}.png"
            img_matches_raw_path: Path = (
                log_path / f"img_matches_raw_{postfix}.png"
            )
            img_matches_ransac_path: Path = (
                log_path / f"img_matches_ransac_{postfix}.png"
            )
            cv2.imwrite(
                str(img_keypoints_path),
                output_keypoints[:, :, ::-1].copy(),  # RGB -> BGR
            )
            cv2.imwrite(
                str(img_matches_raw_path),
                output_matches_raw[:, :, ::-1].copy(),  # RGB -> BGR
            )
            cv2.imwrite(
                str(img_matches_ransac_path),
                output_matches_ransac[:, :, ::-1].copy(),  # RGB -> BGR
            )
            plt.close("all")


if __name__ == "__main__":
    config = load_config(ROOT / "common/config.yaml")
    api = ImageMatchingAPI(config)