Why can't we find the vector Z and have to use W ?

pbaylies / stylegan-encoder

StyleGAN Encoder - converts real images to latent space

Other

740 stars 182 forks source link

Why can't we find the vector Z and have to use W ? #64

Closed datduong closed 3 years ago

datduong commented 3 years ago

Does anyone know why we can't directly optimize for Z which is the 1st input to StyleGAN? Why do we have to optimize for W which is an intermediate input?

Thanks.

pbaylies commented 3 years ago

Historically optimizing for W directly tends to work better, because the StyleGAN mapping network disentangles the image features into latents that are semantically meaningful, whereas the Z input is just a gaussian distribution.

datduong commented 3 years ago

Thanks.