quyan

Results 2 comments of quyan

Hi, I think i have known why there are no BN layers in teacher structure, > "Folded Models below have batch_norm parameters and statistics folded into convolutional layers for speed....