You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Oct 11, 2023. It is now read-only.
Perceptual loss (called feature loss in SwapNet) is key to generating the texture details. Without perceptual loss, the generator can only create clothing shapes without details.
I've currently set the perceptual loss weight to 0 (the --lambda_feat option), because any other value currently causes a lot of noise throughout the entire image. Here's an example.
Perceptual loss (called feature loss in SwapNet) is key to generating the texture details. Without perceptual loss, the generator can only create clothing shapes without details.
I've currently set the perceptual loss weight to 0 (the --lambda_feat option), because any other value currently causes a lot of noise throughout the entire image. Here's an example.
I believe my perceptual loss implementation is buggy. It could be an issue of which layers we compare for perceptual loss.
The text was updated successfully, but these errors were encountered: