Depth result of run_colmap is not same as provided depth data

[weizhenFrank]

Hi, thanks for sharing the code to help us to run a modified version of colmap. I download scannet data and use official python toolkit https://github.com/ScanNet/ScanNet/tree/master/SensReader/python to extract the data.

Then I run the https://github.com/zju3dv/manhattan_sdf/blob/main/docs/run_colmap/run.py.

However, when I check the consistency between your provided data and the result from above script. I found two differences.

1. The name of depth folder is depth_patchmatch rather than depth_colmap in provided data.
2. The depth image has different values between your provied data and the result from above script, such as the screenshot showed.
   
   In this screenshot, pub10 is your provided data for 0050_00 scene of 10th image, obj10 is the result from your script.
   You can see that there's inconsistency.
   But the final metric is quite similar, as the screenshot shows.

Here Provided data is your provided data and From Scannet is from downloaded scannet data.

For the sake of rigorousness, I would like to know whether the difference is normal. I would appreciate a lot for your response!

[ghy0324]

Hi! Thanks for your interest!
The difference of depth folder names is due to some historical reasons, you can just change patchmatch to colmap in run.py. I’m not very sure why the depth values are different, maybe related to the version of COLMAP. But I think it’s OK if you can obtain similar final results.