Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox

Semantic and Geometric Unfolding of StyleGAN Latent Space

Jul 09, 2021
Mustafa Shukor, Xu Yao, Bharath Bhushan Damodaran, Pierre Hellier

Share this with someone who'll enjoy it:

Generative adversarial networks (GANs) have proven to be surprisingly efficient for image editing by inverting and manipulating the latent code corresponding to a natural image. This property emerges from the disentangled nature of the latent space. In this paper, we identify two geometric limitations of such latent space: (a) euclidean distances differ from image perceptual distance, and (b) disentanglement is not optimal and facial attribute separation using linear model is a limiting hypothesis. We thus propose a new method to learn a proxy latent representation using normalizing flows to remedy these limitations, and show that this leads to a more efficient space for face image editing.

* 16 pages 

   Access Paper Source

Share this with someone who'll enjoy it: