/faceswap-autoencoder

Pytorch implementation of a Face Swap Autoencoder

Primary LanguagePython

Faceswap-Autoencoder

This is a Pytorch implementation of a Face Swap Autoencoder, roughly based on Shaonlu's tensorflow implementation..

Notes

  • Both the autoencoder and the discriminator are using spectral normalization
  • Discriminator is being used only as a learned preceptual loss, not a direct adversarial loss
  • Conv2d has been customized to properly use spectral normalization before a pixel-shuffle
  • Downsampling operations have been remove from VGG-Face to provide more detail in perceptual loss

Example Output

Tom Hanks(Young to Old)

Tom Hanks(Old to Young)

Code Usage

Usage instructions found here: user manual page.