search

genforce / idinvert_pytorch

[ECCV 2020] In-Domain GAN Inversion for Real Image Editing (PyTorch code)
https://genforce.github.io/idinvert/
MIT License
407 stars 54 forks source link

Questions about ResBlock implementation #14

Open doutzi opened 3 years ago

doutzi commented 3 years ago

Hi, first of all thanks for the great work!

I just compared the ResBlock implementation with the original tensorflow implementation and I'm a little bit confused. Shouldn't there be an activation layer (e.g. LeakyReLU) after the first convolutional block? And the WScaleLayer used within the ResBlock, does this have the same impact as in the StyleGAN implementation, i.e. enables equalized learning rate if use_wscale=True?

Thank you for the clarification!

zhujiapeng commented 3 years ago

This is the first version of our encoder structure, the weights give here are exactly match the encoder's structure here. We updated the TensorFlow version a little from the first version.