Processing stucks after ~1000 images at BA

[YaroslavShchekaturov]

HI!

Thank you very much for a wonderfull job you did! I've tried to run a reconstruction python ./main.py --config superpoint+lightglue --images images --outs out --strategy sequential --overlap 2 --force and everything is fine when I run it for a dataset with <1000 images. However, I run it for a dataset with 2000 images it takes significantly more time for BA, especially after 1000 images. I tried to lower the quality here "quality": Quality.HIGH -> "quality": Quality.LOW but got 2024-01-04 16:08:21 | [INFO ] Matching features... 2024-01-04 16:08:21 | [INFO ] 3%|█▉ | 114/4545 [00:37<23:58, 3.08it/s] Traceback (most recent call last): File "C:\Users\yaroslav\Desktop\deep-image-matching\main.py", line 405, in main() File "C:\Users\yaroslav\Desktop\deep-image-matching\main.py", line 238, in main match_path = img_matching.match_pairs(feature_path) File "C:\Users\yaroslav\Desktop\deep-image-matching\src\deep_image_matching\image_matching.py", line 311, in match_pairs matches = self._matcher.match( File "C:\Users\yaroslav\Desktop\deep-image-matching\src\deep_image_matching\matchers\matcher_base.py", line 279, in match self._matches = self._match_by_tile( File "C:\Users\yaroslav\Desktop\deep-image-matching\src\deep_image_matching\matchers\matcher_base.py", line 380, in _match_by_tile correspondences = self._match_pairs(feats0_tile, feats1_tile) File "C:\Users\yaroslav\anaconda3\envs\slam\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context return func(*args, **kwargs) File "C:\Users\yaroslav\Desktop\deep-image-matching\src\deep_image_matching\matchers\lightglue.py", line 71, in _match_pairs match_res = self._matcher({"image0": feats0, "image1": feats1}) File "C:\Users\yaroslav\anaconda3\envs\slam\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "C:\Users\yaroslav\anaconda3\envs\slam\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl return forward_call(*args, **kwargs) File "C:\Users\yaroslav\Desktop\deep-image-matching\src\deep_image_matching\thirdparty\LightGlue\lightglue\lightglue.py", line 463, in forward return self._forward(data) File "C:\Users\yaroslav\Desktop\deep-image-matching\src\deep_image_matching\thirdparty\LightGlue\lightglue\lightglue.py", line 529, in _forward if self.check_if_stop(token0[..., :m, :], token1[..., :n, :], i, m + n): File "C:\Users\yaroslav\Desktop\deep-image-matching\src\deep_image_matching\thirdparty\LightGlue\lightglue\lightglue.py", line 613, in check_if_stop confidences = torch.cat([confidences0, confidences1], -1) RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 1 but got size 0 for tensor number 1 in the list.