Spaces:
Running
Running
Realcat
commited on
Commit
•
40c4807
1
Parent(s):
4d4dd90
add: matching API
Browse files- README.md +2 -1
- common/api.py +298 -0
- common/config.yaml +2 -0
- common/utils.py +15 -12
- common/viz.py +10 -6
- test_app_cli.py +16 -130
README.md
CHANGED
@@ -38,7 +38,8 @@ The tool currently supports various popular image matching algorithms, namely:
|
|
38 |
- [x] [RoMa](https://github.com/Vincentqyw/RoMa), CVPR 2024
|
39 |
- [x] [DeDoDe](https://github.com/Parskatt/DeDoDe), 3DV 2024
|
40 |
- [ ] [Mickey](https://github.com/nianticlabs/mickey), CVPR 2024
|
41 |
-
- [
|
|
|
42 |
- [x] [LightGlue](https://github.com/cvg/LightGlue), ICCV 2023
|
43 |
- [x] [DarkFeat](https://github.com/THU-LYJ-Lab/DarkFeat), AAAI 2023
|
44 |
- [ ] [ASTR](https://github.com/ASTR2023/ASTR), CVPR 2023
|
|
|
38 |
- [x] [RoMa](https://github.com/Vincentqyw/RoMa), CVPR 2024
|
39 |
- [x] [DeDoDe](https://github.com/Parskatt/DeDoDe), 3DV 2024
|
40 |
- [ ] [Mickey](https://github.com/nianticlabs/mickey), CVPR 2024
|
41 |
+
- [x] [GIM](https://github.com/xuelunshen/gim), ICLR 2024
|
42 |
+
- [ ] [DUSt3R](https://github.com/naver/dust3r), arXiv 2023
|
43 |
- [x] [LightGlue](https://github.com/cvg/LightGlue), ICCV 2023
|
44 |
- [x] [DarkFeat](https://github.com/THU-LYJ-Lab/DarkFeat), AAAI 2023
|
45 |
- [ ] [ASTR](https://github.com/ASTR2023/ASTR), CVPR 2023
|
common/api.py
ADDED
@@ -0,0 +1,298 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import torch
|
3 |
+
import warnings
|
4 |
+
import numpy as np
|
5 |
+
from pathlib import Path
|
6 |
+
from typing import Dict, Any, Optional, Tuple, List, Union
|
7 |
+
from hloc import logger
|
8 |
+
from hloc import match_dense, match_features, extract_features
|
9 |
+
from hloc.utils.viz import add_text, plot_keypoints
|
10 |
+
from .utils import (
|
11 |
+
load_config,
|
12 |
+
get_model,
|
13 |
+
get_feature_model,
|
14 |
+
filter_matches,
|
15 |
+
device,
|
16 |
+
ROOT,
|
17 |
+
)
|
18 |
+
from .viz import (
|
19 |
+
fig2im,
|
20 |
+
plot_images,
|
21 |
+
display_matches,
|
22 |
+
)
|
23 |
+
import matplotlib.pyplot as plt
|
24 |
+
|
25 |
+
warnings.simplefilter("ignore")
|
26 |
+
|
27 |
+
|
28 |
+
class ImageMatchingAPI(torch.nn.Module):
|
29 |
+
default_conf = {
|
30 |
+
"dense": True,
|
31 |
+
"matcher": {
|
32 |
+
"model": {
|
33 |
+
"name": "topicfm",
|
34 |
+
"match_threshold": 0.2,
|
35 |
+
}
|
36 |
+
},
|
37 |
+
"feature": {
|
38 |
+
"model": {
|
39 |
+
"name": "xfeat",
|
40 |
+
"max_keypoints": 1024,
|
41 |
+
"keypoint_threshold": 0.015,
|
42 |
+
}
|
43 |
+
},
|
44 |
+
"ransac": {
|
45 |
+
"enable": True,
|
46 |
+
"estimator": "poselib",
|
47 |
+
"geometry": "homography",
|
48 |
+
"method": "RANSAC",
|
49 |
+
"reproj_threshold": 3,
|
50 |
+
"confidence": 0.9999,
|
51 |
+
"max_iter": 10000,
|
52 |
+
},
|
53 |
+
}
|
54 |
+
|
55 |
+
def __init__(
|
56 |
+
self,
|
57 |
+
conf: dict = {},
|
58 |
+
device: str = "cpu",
|
59 |
+
detect_threshold: float = 0.015,
|
60 |
+
max_keypoints: int = 1024,
|
61 |
+
match_threshold: float = 0.2,
|
62 |
+
) -> None:
|
63 |
+
"""
|
64 |
+
Initializes an instance of the ImageMatchingAPI class.
|
65 |
+
|
66 |
+
Args:
|
67 |
+
conf (dict): A dictionary containing the configuration parameters.
|
68 |
+
device (str, optional): The device to use for computation. Defaults to "cpu".
|
69 |
+
detect_threshold (float, optional): The threshold for detecting keypoints. Defaults to 0.015.
|
70 |
+
max_keypoints (int, optional): The maximum number of keypoints to extract. Defaults to 1024.
|
71 |
+
match_threshold (float, optional): The threshold for matching keypoints. Defaults to 0.2.
|
72 |
+
|
73 |
+
Returns:
|
74 |
+
None
|
75 |
+
"""
|
76 |
+
super().__init__()
|
77 |
+
self.device = device
|
78 |
+
self.conf = conf = {
|
79 |
+
**self.parse_match_config(self.default_conf),
|
80 |
+
**conf,
|
81 |
+
}
|
82 |
+
self._updata_config(detect_threshold, max_keypoints, match_threshold)
|
83 |
+
self._init_models()
|
84 |
+
self.pred = None
|
85 |
+
|
86 |
+
def parse_match_config(self, conf):
|
87 |
+
if conf["dense"]:
|
88 |
+
return {
|
89 |
+
**conf,
|
90 |
+
"matcher": match_dense.confs.get(
|
91 |
+
conf["matcher"]["model"]["name"]
|
92 |
+
),
|
93 |
+
"dense": True,
|
94 |
+
}
|
95 |
+
else:
|
96 |
+
return {
|
97 |
+
**conf,
|
98 |
+
"feature": extract_features.confs.get(
|
99 |
+
conf["feature"]["model"]["name"]
|
100 |
+
),
|
101 |
+
"matcher": match_features.confs.get(
|
102 |
+
conf["matcher"]["model"]["name"]
|
103 |
+
),
|
104 |
+
"dense": False,
|
105 |
+
}
|
106 |
+
|
107 |
+
def _updata_config(
|
108 |
+
self,
|
109 |
+
detect_threshold: float = 0.015,
|
110 |
+
max_keypoints: int = 1024,
|
111 |
+
match_threshold: float = 0.2,
|
112 |
+
):
|
113 |
+
self.dense = self.conf["dense"]
|
114 |
+
if self.conf["dense"]:
|
115 |
+
self.conf["matcher"]["model"]["match_threshold"] = match_threshold
|
116 |
+
else:
|
117 |
+
self.conf["feature"]["model"]["max_keypoints"] = max_keypoints
|
118 |
+
self.conf["feature"]["model"][
|
119 |
+
"keypoint_threshold"
|
120 |
+
] = detect_threshold
|
121 |
+
self.match_conf = self.conf["matcher"]
|
122 |
+
self.extract_conf = self.conf["feature"]
|
123 |
+
|
124 |
+
def _init_models(self):
|
125 |
+
# initialize matcher
|
126 |
+
self.matcher = get_model(self.conf["matcher"])
|
127 |
+
# initialize extractor
|
128 |
+
if self.dense:
|
129 |
+
self.extractor = None
|
130 |
+
else:
|
131 |
+
self.extractor = get_feature_model(self.conf["feature"])
|
132 |
+
|
133 |
+
def _forward(self, img0, img1):
|
134 |
+
if self.dense:
|
135 |
+
pred = match_dense.match_images(
|
136 |
+
self.matcher,
|
137 |
+
img0,
|
138 |
+
img1,
|
139 |
+
self.match_conf["preprocessing"],
|
140 |
+
device=self.device,
|
141 |
+
)
|
142 |
+
last_fixed = "{}".format(self.match_conf["model"]["name"])
|
143 |
+
else:
|
144 |
+
pred0 = extract_features.extract(
|
145 |
+
self.extractor, img0, self.extract_conf["preprocessing"]
|
146 |
+
)
|
147 |
+
pred1 = extract_features.extract(
|
148 |
+
self.extractor, img1, self.extract_conf["preprocessing"]
|
149 |
+
)
|
150 |
+
pred = match_features.match_images(self.matcher, pred0, pred1)
|
151 |
+
return pred
|
152 |
+
|
153 |
+
@torch.inference_mode()
|
154 |
+
def forward(
|
155 |
+
self,
|
156 |
+
img0: np.ndarray,
|
157 |
+
img1: np.ndarray,
|
158 |
+
) -> Dict[str, np.ndarray]:
|
159 |
+
"""
|
160 |
+
Forward pass of the image matching API.
|
161 |
+
|
162 |
+
Args:
|
163 |
+
img0: A 3D NumPy array of shape (H, W, C) representing the first image.
|
164 |
+
Values are in the range [0, 1] and are in RGB mode.
|
165 |
+
img1: A 3D NumPy array of shape (H, W, C) representing the second image.
|
166 |
+
Values are in the range [0, 1] and are in RGB mode.
|
167 |
+
|
168 |
+
Returns:
|
169 |
+
A dictionary containing the following keys:
|
170 |
+
- image0_orig: The original image 0.
|
171 |
+
- image1_orig: The original image 1.
|
172 |
+
- keypoints0_orig: The keypoints detected in image 0.
|
173 |
+
- keypoints1_orig: The keypoints detected in image 1.
|
174 |
+
- mkeypoints0_orig: The raw matches between image 0 and image 1.
|
175 |
+
- mkeypoints1_orig: The raw matches between image 1 and image 0.
|
176 |
+
- mmkeypoints0_orig: The RANSAC inliers in image 0.
|
177 |
+
- mmkeypoints1_orig: The RANSAC inliers in image 1.
|
178 |
+
- mconf: The confidence scores for the raw matches.
|
179 |
+
- mmconf: The confidence scores for the RANSAC inliers.
|
180 |
+
"""
|
181 |
+
# Take as input a pair of images (not a batch)
|
182 |
+
assert isinstance(img0, np.ndarray)
|
183 |
+
assert isinstance(img1, np.ndarray)
|
184 |
+
self.pred = self._forward(img0, img1)
|
185 |
+
if self.conf["ransac"]["enable"]:
|
186 |
+
self.pred = self._geometry_check(self.pred)
|
187 |
+
return self.pred
|
188 |
+
|
189 |
+
def _geometry_check(
|
190 |
+
self,
|
191 |
+
pred: Dict[str, Any],
|
192 |
+
) -> Dict[str, Any]:
|
193 |
+
"""
|
194 |
+
Filter matches using RANSAC. If keypoints are available, filter by keypoints.
|
195 |
+
If lines are available, filter by lines. If both keypoints and lines are
|
196 |
+
available, filter by keypoints.
|
197 |
+
|
198 |
+
Args:
|
199 |
+
pred (Dict[str, Any]): dict of matches, including original keypoints.
|
200 |
+
See :func:`filter_matches` for the expected keys.
|
201 |
+
|
202 |
+
Returns:
|
203 |
+
Dict[str, Any]: filtered matches
|
204 |
+
"""
|
205 |
+
pred = filter_matches(
|
206 |
+
pred,
|
207 |
+
ransac_method=self.conf["ransac"]["method"],
|
208 |
+
ransac_reproj_threshold=self.conf["ransac"]["reproj_threshold"],
|
209 |
+
ransac_confidence=self.conf["ransac"]["confidence"],
|
210 |
+
ransac_max_iter=self.conf["ransac"]["max_iter"],
|
211 |
+
)
|
212 |
+
return pred
|
213 |
+
|
214 |
+
def visualize(
|
215 |
+
self,
|
216 |
+
log_path: Optional[Path] = None,
|
217 |
+
) -> None:
|
218 |
+
"""
|
219 |
+
Visualize the matches.
|
220 |
+
|
221 |
+
Args:
|
222 |
+
log_path (Path, optional): The directory to save the images. Defaults to None.
|
223 |
+
|
224 |
+
Returns:
|
225 |
+
None
|
226 |
+
"""
|
227 |
+
if self.conf["dense"]:
|
228 |
+
postfix = str(self.conf["matcher"]["model"]["name"])
|
229 |
+
else:
|
230 |
+
postfix = "{}_{}".format(
|
231 |
+
str(self.conf["feature"]["model"]["name"]),
|
232 |
+
str(self.conf["matcher"]["model"]["name"]),
|
233 |
+
)
|
234 |
+
titles = [
|
235 |
+
"Image 0 - Keypoints",
|
236 |
+
"Image 1 - Keypoints",
|
237 |
+
]
|
238 |
+
pred: Dict[str, Any] = self.pred
|
239 |
+
image0: np.ndarray = pred["image0_orig"]
|
240 |
+
image1: np.ndarray = pred["image1_orig"]
|
241 |
+
output_keypoints: np.ndarray = plot_images(
|
242 |
+
[image0, image1], titles=titles, dpi=300
|
243 |
+
)
|
244 |
+
if (
|
245 |
+
"keypoints0_orig" in pred.keys()
|
246 |
+
and "keypoints1_orig" in pred.keys()
|
247 |
+
):
|
248 |
+
plot_keypoints([pred["keypoints0_orig"], pred["keypoints1_orig"]])
|
249 |
+
text: str = (
|
250 |
+
f"# keypoints0: {len(pred['keypoints0_orig'])} \n"
|
251 |
+
+ f"# keypoints1: {len(pred['keypoints1_orig'])}"
|
252 |
+
)
|
253 |
+
add_text(0, text, fs=15)
|
254 |
+
output_keypoints = fig2im(output_keypoints)
|
255 |
+
# plot images with raw matches
|
256 |
+
titles = [
|
257 |
+
"Image 0 - Raw matched keypoints",
|
258 |
+
"Image 1 - Raw matched keypoints",
|
259 |
+
]
|
260 |
+
output_matches_raw, num_matches_raw = display_matches(
|
261 |
+
pred, titles=titles, tag="KPTS_RAW"
|
262 |
+
)
|
263 |
+
# plot images with ransac matches
|
264 |
+
titles = [
|
265 |
+
"Image 0 - Ransac matched keypoints",
|
266 |
+
"Image 1 - Ransac matched keypoints",
|
267 |
+
]
|
268 |
+
output_matches_ransac, num_matches_ransac = display_matches(
|
269 |
+
pred, titles=titles, tag="KPTS_RANSAC"
|
270 |
+
)
|
271 |
+
if log_path is not None:
|
272 |
+
img_keypoints_path: Path = log_path / f"img_keypoints_{postfix}.png"
|
273 |
+
img_matches_raw_path: Path = (
|
274 |
+
log_path / f"img_matches_raw_{postfix}.png"
|
275 |
+
)
|
276 |
+
img_matches_ransac_path: Path = (
|
277 |
+
log_path / f"img_matches_ransac_{postfix}.png"
|
278 |
+
)
|
279 |
+
cv2.imwrite(
|
280 |
+
str(img_keypoints_path),
|
281 |
+
output_keypoints[:, :, ::-1].copy(), # RGB -> BGR
|
282 |
+
)
|
283 |
+
cv2.imwrite(
|
284 |
+
str(img_matches_raw_path),
|
285 |
+
output_matches_raw[:, :, ::-1].copy(), # RGB -> BGR
|
286 |
+
)
|
287 |
+
cv2.imwrite(
|
288 |
+
str(img_matches_ransac_path),
|
289 |
+
output_matches_ransac[:, :, ::-1].copy(), # RGB -> BGR
|
290 |
+
)
|
291 |
+
plt.close("all")
|
292 |
+
|
293 |
+
|
294 |
+
if __name__ == "__main__":
|
295 |
+
import argparse
|
296 |
+
|
297 |
+
config = load_config(ROOT / "common/config.yaml")
|
298 |
+
test_api(config)
|
common/config.yaml
CHANGED
@@ -319,6 +319,7 @@ matcher_zoo:
|
|
319 |
project: null
|
320 |
display: true
|
321 |
gluestick:
|
|
|
322 |
matcher: gluestick
|
323 |
dense: true
|
324 |
info:
|
@@ -329,6 +330,7 @@ matcher_zoo:
|
|
329 |
project: https://iago-suarez.com/gluestick
|
330 |
display: true
|
331 |
sold2:
|
|
|
332 |
matcher: sold2
|
333 |
dense: true
|
334 |
info:
|
|
|
319 |
project: null
|
320 |
display: true
|
321 |
gluestick:
|
322 |
+
enable: false
|
323 |
matcher: gluestick
|
324 |
dense: true
|
325 |
info:
|
|
|
330 |
project: https://iago-suarez.com/gluestick
|
331 |
display: true
|
332 |
sold2:
|
333 |
+
enable: false
|
334 |
matcher: sold2
|
335 |
dense: true
|
336 |
info:
|
common/utils.py
CHANGED
@@ -78,21 +78,24 @@ def get_matcher_zoo(
|
|
78 |
"""
|
79 |
matcher_zoo_restored = {}
|
80 |
for k, v in matcher_zoo.items():
|
81 |
-
|
82 |
-
if dense:
|
83 |
-
matcher_zoo_restored[k] = {
|
84 |
-
"matcher": match_dense.confs.get(v["matcher"]),
|
85 |
-
"dense": dense,
|
86 |
-
}
|
87 |
-
else:
|
88 |
-
matcher_zoo_restored[k] = {
|
89 |
-
"feature": extract_features.confs.get(v["feature"]),
|
90 |
-
"matcher": match_features.confs.get(v["matcher"]),
|
91 |
-
"dense": dense,
|
92 |
-
}
|
93 |
return matcher_zoo_restored
|
94 |
|
95 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
def get_model(match_conf: Dict[str, Any]):
|
97 |
"""
|
98 |
Load a matcher model from the provided configuration.
|
|
|
78 |
"""
|
79 |
matcher_zoo_restored = {}
|
80 |
for k, v in matcher_zoo.items():
|
81 |
+
matcher_zoo_restored[k] = parse_match_config(v)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
82 |
return matcher_zoo_restored
|
83 |
|
84 |
|
85 |
+
def parse_match_config(conf):
|
86 |
+
if conf["dense"]:
|
87 |
+
return {
|
88 |
+
"matcher": match_dense.confs.get(conf["matcher"]),
|
89 |
+
"dense": True,
|
90 |
+
}
|
91 |
+
else:
|
92 |
+
return {
|
93 |
+
"feature": extract_features.confs.get(conf["feature"]),
|
94 |
+
"matcher": match_features.confs.get(conf["matcher"]),
|
95 |
+
"dense": False,
|
96 |
+
}
|
97 |
+
|
98 |
+
|
99 |
def get_model(match_conf: Dict[str, Any]):
|
100 |
"""
|
101 |
Load a matcher model from the provided configuration.
|
common/viz.py
CHANGED
@@ -415,12 +415,17 @@ def display_matches(
|
|
415 |
num_inliers = 0
|
416 |
KPTS0_KEY = None
|
417 |
KPTS1_KEY = None
|
|
|
418 |
if tag == "KPTS_RAW":
|
419 |
KPTS0_KEY = "mkeypoints0_orig"
|
420 |
KPTS1_KEY = "mkeypoints1_orig"
|
|
|
|
|
421 |
elif tag == "KPTS_RANSAC":
|
422 |
KPTS0_KEY = "mmkeypoints0_orig"
|
423 |
KPTS1_KEY = "mmkeypoints1_orig"
|
|
|
|
|
424 |
else:
|
425 |
# TODO: LINES_RAW, LINES_RANSAC
|
426 |
raise ValueError(f"Unknown tag: {tag}")
|
@@ -434,16 +439,14 @@ def display_matches(
|
|
434 |
mkpts0 = pred[KPTS0_KEY]
|
435 |
mkpts1 = pred[KPTS1_KEY]
|
436 |
num_inliers = len(mkpts0)
|
437 |
-
if
|
438 |
-
|
439 |
-
else:
|
440 |
-
mmconf = np.ones(len(mkpts0))
|
441 |
fig_mkpts = draw_matches_core(
|
442 |
mkpts0,
|
443 |
mkpts1,
|
444 |
img0,
|
445 |
img1,
|
446 |
-
|
447 |
dpi=dpi,
|
448 |
titles=titles,
|
449 |
texts=texts,
|
@@ -472,7 +475,8 @@ def display_matches(
|
|
472 |
# keypoints
|
473 |
mkpts0 = pred.get("line_keypoints0_orig")
|
474 |
mkpts1 = pred.get("line_keypoints1_orig")
|
475 |
-
|
|
|
476 |
if mkpts0 is not None and mkpts1 is not None:
|
477 |
num_inliers = len(mkpts0)
|
478 |
if "mconf" in pred:
|
|
|
415 |
num_inliers = 0
|
416 |
KPTS0_KEY = None
|
417 |
KPTS1_KEY = None
|
418 |
+
confid = None
|
419 |
if tag == "KPTS_RAW":
|
420 |
KPTS0_KEY = "mkeypoints0_orig"
|
421 |
KPTS1_KEY = "mkeypoints1_orig"
|
422 |
+
if "mconf" in pred:
|
423 |
+
confid = pred["mconf"]
|
424 |
elif tag == "KPTS_RANSAC":
|
425 |
KPTS0_KEY = "mmkeypoints0_orig"
|
426 |
KPTS1_KEY = "mmkeypoints1_orig"
|
427 |
+
if "mmconf" in pred:
|
428 |
+
confid = pred["mmconf"]
|
429 |
else:
|
430 |
# TODO: LINES_RAW, LINES_RANSAC
|
431 |
raise ValueError(f"Unknown tag: {tag}")
|
|
|
439 |
mkpts0 = pred[KPTS0_KEY]
|
440 |
mkpts1 = pred[KPTS1_KEY]
|
441 |
num_inliers = len(mkpts0)
|
442 |
+
if confid is None:
|
443 |
+
confid = np.ones(len(mkpts0))
|
|
|
|
|
444 |
fig_mkpts = draw_matches_core(
|
445 |
mkpts0,
|
446 |
mkpts1,
|
447 |
img0,
|
448 |
img1,
|
449 |
+
confid,
|
450 |
dpi=dpi,
|
451 |
titles=titles,
|
452 |
texts=texts,
|
|
|
475 |
# keypoints
|
476 |
mkpts0 = pred.get("line_keypoints0_orig")
|
477 |
mkpts1 = pred.get("line_keypoints1_orig")
|
478 |
+
fig = None
|
479 |
+
breakpoint()
|
480 |
if mkpts0 is not None and mkpts1 is not None:
|
481 |
num_inliers = len(mkpts0)
|
482 |
if "mconf" in pred:
|
test_app_cli.py
CHANGED
@@ -1,153 +1,39 @@
|
|
1 |
import cv2
|
2 |
import warnings
|
|
|
3 |
from pathlib import Path
|
4 |
from hloc import logger
|
5 |
-
from hloc import matchers, extractors, logger
|
6 |
-
from hloc import match_dense, match_features, extract_features
|
7 |
-
from hloc.utils.viz import add_text, plot_keypoints
|
8 |
from common.utils import (
|
9 |
-
load_config,
|
10 |
-
get_model,
|
11 |
-
get_feature_model,
|
12 |
-
ransac_zoo,
|
13 |
get_matcher_zoo,
|
14 |
-
|
15 |
device,
|
16 |
ROOT,
|
17 |
)
|
18 |
-
from common.
|
19 |
-
fig2im,
|
20 |
-
plot_images,
|
21 |
-
display_matches,
|
22 |
-
plot_color_line_matches,
|
23 |
-
)
|
24 |
-
import time
|
25 |
-
import matplotlib.pyplot as plt
|
26 |
-
|
27 |
-
warnings.simplefilter("ignore")
|
28 |
|
29 |
-
|
30 |
-
def test_modules(config: dict):
|
31 |
img_path1 = ROOT / "datasets/sacre_coeur/mapping/02928139_3448003521.jpg"
|
32 |
img_path2 = ROOT / "datasets/sacre_coeur/mapping/17295357_9106075285.jpg"
|
33 |
-
image0 = cv2.imread(str(img_path1))
|
34 |
-
image1 = cv2.imread(str(img_path2))
|
35 |
-
keypoint_threshold = 0.0
|
36 |
-
extract_max_keypoints = 2000
|
37 |
-
match_threshold = 0.2
|
38 |
-
log_path = ROOT / "experiments"
|
39 |
-
log_path.mkdir(exist_ok=True, parents=True)
|
40 |
|
41 |
matcher_zoo_restored = get_matcher_zoo(config["matcher_zoo"])
|
42 |
for k, v in matcher_zoo_restored.items():
|
43 |
if image0 is None or image1 is None:
|
44 |
logger.error("Error: No images found! Please upload two images.")
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
match_conf["model"]["max_keypoints"] = extract_max_keypoints
|
54 |
-
matcher = get_model(match_conf)
|
55 |
-
t1 = time.time()
|
56 |
-
if v["dense"]:
|
57 |
-
pred = match_dense.match_images(
|
58 |
-
matcher,
|
59 |
-
image0,
|
60 |
-
image1,
|
61 |
-
match_conf["preprocessing"],
|
62 |
-
device=device,
|
63 |
-
)
|
64 |
-
del matcher
|
65 |
-
extract_conf = None
|
66 |
-
last_fixed = "{}".format(match_conf["model"]["name"])
|
67 |
else:
|
68 |
-
|
69 |
-
|
70 |
-
# update extract config
|
71 |
-
extract_conf["model"]["max_keypoints"] = extract_max_keypoints
|
72 |
-
extract_conf["model"]["keypoint_threshold"] = keypoint_threshold
|
73 |
-
extractor = get_feature_model(extract_conf)
|
74 |
-
pred0 = extract_features.extract(
|
75 |
-
extractor, image0, extract_conf["preprocessing"]
|
76 |
-
)
|
77 |
-
pred1 = extract_features.extract(
|
78 |
-
extractor, image1, extract_conf["preprocessing"]
|
79 |
-
)
|
80 |
-
pred = match_features.match_images(matcher, pred0, pred1)
|
81 |
-
del extractor
|
82 |
-
last_fixed = "{}_{}".format(
|
83 |
-
extract_conf["model"]["name"], match_conf["model"]["name"]
|
84 |
-
)
|
85 |
-
|
86 |
-
# keypoints on images
|
87 |
-
logger.info(f"Match features done using: {time.time()-t1:.3f}s")
|
88 |
-
t1 = time.time()
|
89 |
-
texts = [
|
90 |
-
f"image pairs: {img_path1.name} & {img_path2.name}",
|
91 |
-
"",
|
92 |
-
]
|
93 |
-
titles = [
|
94 |
-
"Image 0 - Keypoints",
|
95 |
-
"Image 1 - Keypoints",
|
96 |
-
]
|
97 |
-
output_keypoints = plot_images([image0, image1], titles=titles, dpi=300)
|
98 |
-
if "keypoints0" in pred.keys() and "keypoints1" in pred.keys():
|
99 |
-
plot_keypoints([pred["keypoints0"], pred["keypoints1"]])
|
100 |
-
text = (
|
101 |
-
f"# keypoints0: {len(pred['keypoints0'])} \n"
|
102 |
-
+ f"# keypoints1: {len(pred['keypoints1'])}"
|
103 |
-
)
|
104 |
-
add_text(0, text, fs=15)
|
105 |
-
output_keypoints = fig2im(output_keypoints)
|
106 |
-
|
107 |
-
# plot images with raw matches
|
108 |
-
titles = [
|
109 |
-
"Image 0 - Raw matched keypoints",
|
110 |
-
"Image 1 - Raw matched keypoints",
|
111 |
-
]
|
112 |
-
output_matches_raw, num_matches_raw = display_matches(
|
113 |
-
pred, titles=titles
|
114 |
-
)
|
115 |
-
logger.info(f"Plot keypoints done using: {time.time()-t1:.3f}s")
|
116 |
-
t1 = time.time()
|
117 |
-
|
118 |
-
filter_matches(
|
119 |
-
pred,
|
120 |
-
ransac_method=config["defaults"]["ransac_method"],
|
121 |
-
ransac_reproj_threshold=config["defaults"][
|
122 |
-
"ransac_reproj_threshold"
|
123 |
-
],
|
124 |
-
ransac_confidence=config["defaults"]["ransac_confidence"],
|
125 |
-
ransac_max_iter=config["defaults"]["ransac_max_iter"],
|
126 |
-
)
|
127 |
-
# plot images with ransac matches
|
128 |
-
titles = [
|
129 |
-
"Image 0 - Ransac matched keypoints",
|
130 |
-
"Image 1 - Ransac matched keypoints",
|
131 |
-
]
|
132 |
-
output_matches_ransac, num_matches_ransac = display_matches(
|
133 |
-
pred, titles=titles
|
134 |
-
)
|
135 |
-
logger.info(f"RANSAC matches done using: {time.time()-t1:.3f}s")
|
136 |
-
|
137 |
-
img_keypoints_path = log_path / f"img_keypoints_{last_fixed}.png"
|
138 |
-
img_matches_raw_path = log_path / f"img_matches_raw_{last_fixed}.png"
|
139 |
-
img_matches_ransac_path = (
|
140 |
-
log_path / f"img_matches_ransac_{last_fixed}.png"
|
141 |
-
)
|
142 |
-
cv2.imwrite(str(img_keypoints_path), output_keypoints)
|
143 |
-
cv2.imwrite(str(img_matches_raw_path), output_matches_raw)
|
144 |
-
cv2.imwrite(str(img_matches_ransac_path), output_matches_ransac)
|
145 |
-
|
146 |
-
plt.close("all")
|
147 |
-
|
148 |
|
149 |
if __name__ == "__main__":
|
150 |
import argparse
|
151 |
|
152 |
config = load_config(ROOT / "common/config.yaml")
|
153 |
-
|
|
|
1 |
import cv2
|
2 |
import warnings
|
3 |
+
import numpy as np
|
4 |
from pathlib import Path
|
5 |
from hloc import logger
|
|
|
|
|
|
|
6 |
from common.utils import (
|
|
|
|
|
|
|
|
|
7 |
get_matcher_zoo,
|
8 |
+
load_config,
|
9 |
device,
|
10 |
ROOT,
|
11 |
)
|
12 |
+
from common.api import ImageMatchingAPI
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
+
def test_api(config: dict = None):
|
|
|
15 |
img_path1 = ROOT / "datasets/sacre_coeur/mapping/02928139_3448003521.jpg"
|
16 |
img_path2 = ROOT / "datasets/sacre_coeur/mapping/17295357_9106075285.jpg"
|
17 |
+
image0 = cv2.imread(str(img_path1))[:, :, ::-1]
|
18 |
+
image1 = cv2.imread(str(img_path2))[:, :, ::-1]
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
matcher_zoo_restored = get_matcher_zoo(config["matcher_zoo"])
|
21 |
for k, v in matcher_zoo_restored.items():
|
22 |
if image0 is None or image1 is None:
|
23 |
logger.error("Error: No images found! Please upload two images.")
|
24 |
+
enable = config["matcher_zoo"][k].get("enable", True)
|
25 |
+
if enable:
|
26 |
+
logger.info(f"Testing {k} ...")
|
27 |
+
api = ImageMatchingAPI(conf=v, device=device)
|
28 |
+
api(image0, image1)
|
29 |
+
log_path = ROOT / "experiments1"
|
30 |
+
log_path.mkdir(exist_ok=True, parents=True)
|
31 |
+
api.visualize(log_path=log_path)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
else:
|
33 |
+
logger.info(f"Skipping {k} ...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
if __name__ == "__main__":
|
36 |
import argparse
|
37 |
|
38 |
config = load_config(ROOT / "common/config.yaml")
|
39 |
+
test_api(config)
|