search

zengwang430521 / DecoMR

Repository for the paper " 3D Human Mesh Regression with Dense Correspondence "
Apache License 2.0
168 stars 23 forks source link

what is the meaning of tv_loss? #22

Closed wangzheallen closed 3 years ago

wangzheallen commented 3 years ago

definition of tv_loss: https://github.com/zengwang430521/DecoMR/blob/69b5986d8545097cf607bd9e6f2f836de7434ce5/train/base_trainer.py#L246 it is with weight 0.0001 when training on up-3d. I am not sure about the meaning of the loss.

Will this loss only apply only at uv_res 128 or less? If I have a uv_res at 256, does that mean we can get rid of the loss?

zengwang430521 commented 3 years ago

Hi, tv_loss is only a total variation regularizer. This loss item has very little effect to the performance. We add this loss item to make the estimated location map look better in unsupervised area.