rosinality / style-based-gan-pytorch

Implementation A Style-Based Generator Architecture for Generative Adversarial Networks in PyTorch
Other
1.1k stars 232 forks source link

gen_in1 and gen_in2 #99

Open KumapowerLIU opened 4 years ago

KumapowerLIU commented 4 years ago

Hi, I think tihis a great work. But I have a quitions about the gen_in1 and gen_in2 , when traning the discriminator, the latent feature is gen_in1 while when traning the generator the latent feature is gen_in2, the fake image in these two steps will different, is this some trick or may be mentioned in the paper. Thank you.

rosinality commented 4 years ago

Actually it is just a matter of the choice - that is, backprop with samples that used to update discriminators or with new samples.