domain translation using latents

[Axelsalamartin]

Is it possible to easily translate a source image from one domain to another using latent and not ref images?

I see a fucntion translate_using_latent(nets, args, x_src, y_trg_list, z_trg_list, psi, filename): in utils.py (line 78) but it is never used and I am not sure of how "y_trg_list" and "z_trg_list" are supposed to be.

[Axelsalamartin]

In the end I managed to make it work easily by using the translate_using_latent() located in utils.py :

I used it in the sample function located in Solver.py (line 174).
I replaced :

• utils.translate_using_reference(nets_ema, args, src.x, ref.x, ref.y, fname) (line 186)
  by :
• utils.translate_using_latent(nets_ema, args, src.x, [torch.ones(src.x.size(0)).long().to('cuda')],[torch.randn(src.x.size(0), 16).to('cuda')], .5, fname)

so [torch.ones(src.x.size(0)).long().to('cuda')] is the y_target list, here it is all ones so all source images will be change to the domain 1 which is 'male' in the case of the celeb A pretrained network.

[torch.randn(src.x.size(0), 16).to('cuda')] is the list of latent vector that'll be used in the mapping_network alongside the y_target list to get the styles which will be used in the generator with the source images to generate the fake images.

The best way would be to then create a function sample_with_latent in Solver.py but this was just to make it work really quickly :)