ewrfcas/MVSFormer

What are the thresholds for depth confidence filter?

AlbertHuyb opened this issue · 1 comments

Thanks for the nice work!

As mentioned in the paper, the depth maps are filtered by certain thresholds.

Specifically, how should I set the prob_threshold parameters in depthfusion.py for the tnt intermediate and advanced dataset to reproduce results in the paper?

Moreover, could you please provide the post-processing steps that prepare the submission to TNT official evaluation?

Thanks. You could follow the instruction of README to reproduce the results.
prob_threshold/conf should be set with 0.5.
You could follow MVSNet to calculate the view similarity for more source views, but the performance gap is not too large.