Open youkaichao opened 6 years ago
Do you mean you want to train models with the loss proposed in the original GANs paper? If so, you can do that by setting loss_type to 'dcgan' in our code: https://github.com/pfnet-research/sngan_projection/blob/master/updater.py#L37
the output of inner-product is a scalar. In current implementation, this repo uses hinge loss without sigmoid activation.
So I wonder how to use projection in models with sigmoid activation. Can you give some suggestions?