The mask is unable to remove the background during reconstruction

[gnulife]

I have already generated the corresponding mask files in the data/masks/ directory by running python preprocess_mask.py --data <path to data>. Then, I used python train.py -s <path to data> -m output/trained_result --w_mask for the reconstruction. However, the resulting point_cloud.ply still contains the background. There were no error messages throughout the process. Could you please tell me what the issue might be?

By the way, there is a mistake in the documentation; there should be two hyphens before the w_mask parameter, just like "--w_mask"

[solomon-ma]

Could you show some results after " preprocess_mask.py --data " ?

[gnulife]

[hugoycj]

@gnulife We just updated gaustudio dataloader to support mask input yesterday. You may need to pull and install the latest version gaustudio to support mask supervision.

[gnulife]

I performed a git pull before running python setup.py develop, and when I checked the official website, I only saw updates from three days ago; there were no updates from yesterday.

[hugoycj]

Sorry for that. It’s my fault. We have not finished the merged yet due to some under-developing feature. We have push a temp branch for testing https://github.com/GAP-LAB-CUHK-SZ/gaustudio/tree/feat/better-colmap-loader

[gnulife]

Thank you for your reply so guickly, I will give it a try again. thank you!

[gnulife]

The effect is very poor. Is it because the background removal effect of Rembg is not good? Do you have any suggestions for improving the condition of the edges?

[gnulife]

When I do not remove the background, the effect is good.What could possibly be the issue?