rosinality/style-based-gan-pytorch

Discriminator missing an activation function after the first convolutional layer

adambielski opened this issue · 2 comments

After from_rgb layer in the discriminator, there is no activation function. A leaky relu nonlinearity should be applied there.

out = self.from_rgb[index](input)

Thank you, I missed it. I will update the repo.

Added this as an option at b160dbc for checkpoint compatibility. I will update checkpoint later.