facebookresearch / robust_cvd

Robust Consistent Video Depth Estimation

Home Page:https://robust-cvd.github.io/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

contrast loss

lkosh opened this issue · comments

Hello
I have noticed that there is an additional loss used by default in your algorithm, yet it wasn't described in the paper. It is defined in loss/contrast_loss.py andit can be disabled by setting "--lambda_contrast_loss 0". If I disable this loss, the final result appears to be extremely blurry.
Could someone explain where this loss comes from and what it does? Was it taken from some other project?
Thank you