Closed ucalyptus2 closed 3 years ago
Thanks for the input @Qualeafclover but the E and F you mention there are from the original StyleGAN paper. In StyleGAN2 the difference between E and F is just that F is a larger model. It has double the number of channels at the higher resolution layers. See this table in the StyleGAN2 paper:
Thanks @justinpinkney
The differences are basically
E: Add noise inputs F: Mixing regularization
For more info, you can find it in the stylegan paper NVIDIA published. https://arxiv.org/pdf/1812.04948.pdf
Note that there are 2 more papers for the improvement of stylegan. https://arxiv.org/pdf/1912.04958.pdf https://arxiv.org/pdf/2006.06676.pdf