search

ForMyCat / SparseGS

Codebase for SparseGS paper
Other
96 stars 11 forks source link

Why are Pearson loss and reg loss mutually exclusive? #21

Open Master-cai opened 3 weeks ago

Master-cai commented 3 weeks ago

Hiya! Thanks for your wonderful work! but i'm confused why are Pearson loss and reg loss mutually exclusive? See https://github.com/ForMyCat/SparseGS/blob/95e7aef29c5562400d3b2b38cc7e90436a432b7c/train.py#L149 and https://github.com/ForMyCat/SparseGS/blob/95e7aef29c5562400d3b2b38cc7e90436a432b7c/train.py#L153

cc @ForMyCat

ForMyCat commented 2 weeks ago

Hi, thank you for you interest in our work.

Because we don't have GT images for warped views, hence no depth maps for pearson loss. I see your point is that we can run DPT model on the warped image (pseudo-GT) and apply pearson loss, but the result is heavily dependent on the warping quality. For RGB loss this is not too much of an issue since we can apply the warping mask. But DPT model cannot take in mask.