Skip to content

Deceptrax123/Style-GAN-Implementation

Repository files navigation

Style GAN

Implemented the Style GAN Architecture as proposed in this paper.

Details of Architecture

Architecture

  • It consists of a mapping network $f$ to learn a disentangled latent space. A disentangled latent space enables the possibility to find direction vectors that correspond to indvidual factors of variation.
  • The synthesis netwok $g$ performs style-mixing. Further, we add random Gaussian Noise to each block as a means to generate stochastic detail.
  • Adaptive Instance Normalization is used in each block due to its efficiency and compact representation.