Issue with convert.py and Missing Sparse Directories in WSL

[EkSulfur]

Hello, I'd encountered an issue when running the convert.py script with Colmap. It seems that the sparse/0 directory is not the final result for the Colmap mapper, and solved it by modifying convert.py to set the input_path to the final sparse directory, as mentioned in #964 (comment)

This change worked for me in the past, but something strange happened recently.

Issue Details:

• On some datasets, there are multiple directories under /distorted/sparse, and I could confirm this was true.

• However, when I tried running convert.py in WSL with the same dataset and Colmap version (3.11.1), the other directories (except /0) mysteriously disappeared, leaving only /0 with just tens of KB of data.

• Additionally, when I ran convert.py in Windows, the same issue occurred: only the /0 directory was left under /distorted/sparse.

• I was able to resolve the issue on Windows by reinstalling Colmap, but this didn't work in WSL. I also removed the Windows environment path in WSL, but the issue persists.

• I attempted to replicate the situation by running convert.py again on Windows, but the issue happened again, and reinstalling Colmap didn’t fix it this time. This is very strange.

Colmap Log:

I suppose the issue is related to the Bundle Adjustment step, where there're lots of repeating no good initial image pair:

I0216 12:00:25.569139 1055 incremental_pipeline.cc:282] Finding good initial image pair
I0216 12:00:25.637390 1055 incremental_pipeline.cc:286] => No good initial image pair found.
...

For the same dataset and Colmap version, this problem (sometimes) doesn't happen on Windows, and it also takes longer to run convert.py there.

I’m unsure why this issue is happening, and would appreciate any help in troubleshooting or resolving it. Thank you!

[ali-john]

Same issue here. I have a multi camera dataset and seems like colmap throws warning "No good initial image pair found"