Does nn.utils.weight_norm apply mean-only batch norm too? I am trying to do semi-supervised learning with GANs on the Cifar-10 dataset and I’ve come across numerous resources that recommend applying weight_norm with mean-only batch norm for the task as it improves performance.
Additionally, I am facing issues when I try to deepcopy a network that uses nn.utils.weight_norm. I get the following error:
RuntimeError: Only Tensors created explicitly by the user (graph leaves) support the deepcopy protocol at the moment
Could you please advise a workaround this issue? I assume this is because, deepcopy cannot copy hooks of a module.