xingmimfl/Deep3DFacereconstruction_pytorch3d

"mean face" problem

HoiM opened this issue · 2 comments

HoiM commented

I found that in your implementation, the first layer of ResNet-50 are frozen. (you set require_grad as False), may I know the reason why you have such a setting?

HoiM commented

Problem found. Not an overfitting but a bug in my test code.

zzhat0706 commented

@HoiM Sorry for bothering，I also found the mean shape problem in training process. I am wondering if you sovled this problem by set require_grad as True，to let even the first layer of Resnet trainable？Thanks.