search

dunbar12138 / DSNeRF

Code release for DS-NeRF (Depth-supervised Neural Radiance Fields)
https://www.cs.cmu.edu/~dsnerf/
MIT License
746 stars 126 forks source link

what does relative loss mean? #88

Open silver-obelisk opened 1 year ago

silver-obelisk commented 1 year ago

Hi,I found you add relative_loss in your code. I tried relative_loss,weighted_loss,and nomal mse, then I foud relative_loss is best but I donnot know why. I wonder what does relative loss mean? Why it better then nomal mse?

dunbar12138 commented 1 year ago

Hi, thanks for your interest.

Relative loss means we normalize the depth error by the depth values so that we take into account the scale of depths, which could vary widely depending on the scene and the imaging setup.