SWAGAN: A Style-based Wavelet-driven Generative Model

  • 2021-02-11 16:43:10
  • Rinon Gal, Dana Cohen, Amit Bermano, Daniel Cohen-Or
  • 48

Abstract

In recent years, considerable progress has been made in the visual quality ofGenerative Adversarial Networks (GANs). Even so, these networks still sufferfrom degradation in quality for high-frequency content, stemming from aspectrally biased architecture, and similarly unfavorable loss functions. Toaddress this issue, we present a novel general-purpose Style and WAvelet basedGAN (SWAGAN) that implements progressive generation in the frequency domain.SWAGAN incorporates wavelets throughout its generator and discriminatorarchitectures, enforcing a frequency-aware latent representation at every stepof the way. This approach yields enhancements in the visual quality of thegenerated images, and considerably increases computational performance. Wedemonstrate the advantage of our method by integrating it into the SyleGAN2framework, and verifying that content generation in the wavelet domain leads tohigher quality images with more realistic high-frequency content. Furthermore,we verify that our model's latent space retains the qualities that allowStyleGAN to serve as a basis for a multitude of editing tasks, and show thatour frequency-aware approach also induces improved downstream visual quality.

 

Quick Read (beta)

loading the full paper ...