genforce / lia Goto Github PK

Hi,
It's not clear if the same training/testing data can be used for decoder and encoder training. Decoder training is exactly the same as StyleGAN as described. This stage can use the 9 ffhq tfrecords from StyleGAN having a resolution of 1024x1024. The encoder training seems to using a data set of resolution 128x128. Where is the data set?

Hello.

Hope you are doing well.

Thank you for your work. It is very impressive. I just wanted to ask you about the contents of the objects stored in the pkl file. I used your encoder and decoder. I got a pkl file that contains G, D, Gs, NE only from the decoder and encoder as well. When I want to use reconstruction.py the following code will not work: E, _, _, Gs, _ = load_pkl(args.restore_path). Do you have an idea of why that might be the case? Why was the E not stored? I believe E is the encoder from real image to the latent 512.

Looking forward to your reply.

Thank you

Hi, thank you for this great work! @zhujiapeng

I would like to know if there is a straightforward way to get the latent code mapped from the real input image to the z/w space using the well-trained encoder, any insights would be highly appreciated!