search

nv-tlabs / semanticGAN_code

Official repo for SemanticGAN https://nv-tlabs.github.io/semanticGAN/
MIT License
180 stars 34 forks source link

There is no softmax in the forward function of GeneratorSeg #12

Closed sangyun884 closed 2 years ago

sangyun884 commented 3 years ago

There is no softmax in the forward function of GeneratorSeg, and the output segmentation map of the generator is directly fed to the discriminator. How does a generator fool a discriminator if the output segmentation map is not normalized?

lidaiqing commented 2 years ago

Hi, sorry for the late reply. It is the same as the image generation branch where there is also no tanh layer. Stylegan is well normalized in each layer, but you can also try using normalization layer before the output.