Jianyuan Wang

For each category, it's the mean size of all objects among the training set. It can also work well if being set as [1,1,1].

not familiar with windows but it looks a certain library (lz4) is not installed or linked to the right path

you can check this document: https://colmap.github.io/format.html What you want is saved in points3D.txt and images.txt

It looks the correspondences (2D matches) are noisy and cannot lead to a reasonable solution. You can check libraries like [hloc](https://github.com/cvg/Hierarchical-Localization) about how to use better detecting and matching methods...

Hi the best option would be generating such an image by youself. You can get the pixel locations of the keypoints for each frame in images.txt (see https://colmap.github.io/format.html#images-txt). Then it...

You could try to slightly lift the camera a bit and put the camera farer. Check how did they do for Co3D, which shares a similar setting to yours as...

My wild guess is probably the images are too huge. 5000x3500 may be too huge, which provides too many keypoints to colmap and hence it takes too much time/out of...

Hey given this it may be not due to the image size. You can use LOG(INFO) to print the logs inside COLMAP, so that you can know which function gets...

I could verify the observation of @Parskatt , where changing THREADS_PER_BLOCK (and kMaxPatchMatchWindowRadius) from 32 to 96 can accelerate it by ~2.8 times on one A100 GPU (although I don't...

@Parskatt after a double check, it seems the results are incorrect, because if I run pycolmap.stereo_fusion, the returned would be: W20240430 14:41:27.907872 1927978 fusion.cc:335] Could not fuse any points. This...