How to return latents from MobileStyleGAN like those in StyleGAN2?

[RahulBhalley]

Hello @bes-dev,

How can I return latents from MobileStyleGAN just like StyleGAN2 does here as following:

if return_latents:
    return image, latent

How can I get the latent variable similar to this in StyleGAN2? Looks like there's some manipulation involved.

[bes-dev]

@RahulBhalley

MobileStyleGAN.pytorch/core/distiller.py

Line 184 in a9776ff

style = self.mapping_net(var)

[RahulBhalley]

Hmm, I used these but the reconstruction results of ReStyle encoder started to become same for any face at around 12k iterations.

Could it have to do anything with the internal manipulation of latent before returning them as in StyleGAN2? Or have I misunderstood something?

[bes-dev]

We didn't test our model with ReStyle

[RahulBhalley]

No problem, I solved the problem. 🙂