search

HuangYG123 / CurricularFace

CurricularFace(CVPR2020)
MIT License
526 stars 72 forks source link

batch_norm parameters issue #15

Open konioyxgq opened 4 years ago

konioyxgq commented 4 years ago

On line 104 of train.py, you write "separate batch_norm parameters from others; do not do weight decay for batch_norm parameters to improve the generalizability"https://github.com/HuangYG123/CurricularFace/blob/8b2f47318117995aa05490c05b455b113489917e/train.py#L104 Is this the conclusion you got from the experiment? Or is the conclusion of a certain paper? Can you explain the reason for this?

HuangYG123 commented 4 years ago

We follow this setting as https://github.com/ZhaoJ9014/face.evoLVe.PyTorch.

konioyxgq commented 4 years ago

Okay thank you. What do you think is the reason for this?