use generator model "Gs.pth" to train pixel2style2pixel encoder

[t109598032]

Hello, thank you for providing this code.
I have the question about how to use the trained StyleGan2 model "Gs.pth" as the decoder of pixel2style2pixel?
(pixel2style2pixel is the architecture combined with encoder+decoder, encoder used to get latent code of picture, decoder is the StyleGan generator, used to generate picture.)

The same issue i had asked here(eladrich/pixel2style2pixel#245 (comment))
The author of pixel2style2pixel has reply me. But it seems that I should train styleGan generator with the"StyleGan-ADA-Pytorch" version firstly, it will generate the .pkl file as the generator, then use its code to convert the .pkl file to .pt file, finally can be used to train pixel2style2pixel encoder.

But I don't want to train the "StyleGan-ADA-Pytorch" again because i already have the "Gs.pth", and training takes me a lot of times.
Is there any solution to use "Gs.pth" as the pre-trained generator(decoder) of pixel2style2pixel?

[adriansahlman]

Hello!

Sorry for the slow reply. To be honest I actually have no idea because I basically stopped working with GANs after creating this repo. So I have no knowledge of pixel2style2pixel and how it works, sorry :(