pytorch-multiple-style-transfer
pytorch-multiple-style-transfer copied to clipboard
Several normalization layers are not conditioned
https://github.com/kewellcjj/pytorch-multiple-style-transfer/blob/a295854382b45076c3f3c334dd6f88e50014fe1a/transformer_net.py#L82
In the original TF implementation the instancenorm layers in the residual blocks are also conditioned as far as I understand.
Leaving them unconditioned in current implementation may lead to lesser quality of the stylization.