Generative Adversarial Networks via a Composite Annealing of Noise and Diffusion

TL;DR

This paper introduces a novel data representation called noisy scale-space (NSS) that stabilizes GAN training by recursively applying smoothing and noise, leading to improved performance over state-of-the-art methods.

Contribution

The paper proposes NSS, a new recursive smoothing and noise injection technique, to enhance GAN training stability and quality, outperforming existing methods on benchmark datasets.

Findings

NSS-based GANs outperform state-of-the-art in most benchmark tests.

NSS stabilizes GAN training by balancing noise and smoothing.

Coarse-to-fine training improves high-frequency detail learning.

Abstract

Generative adversarial network (GAN) is a framework for generating fake data using a set of real examples. However, GAN is unstable in the training stage. In order to stabilize GANs, the noise injection has been used to enlarge the overlap of the real and fake distributions at the cost of increasing variance. The diffusion (or smoothing) may reduce the intrinsic underlying dimensionality of data but it suppresses the capability of GANs to learn high-frequency information in the training procedure. Based on these observations, we propose a data representation for the GAN training, called noisy scale-space (NSS), that recursively applies the smoothing with a balanced noise to data in order to replace the high-frequency information by random data, leading to a coarse-to-fine training of GANs. We experiment with NSS using DCGAN and StyleGAN2 based on benchmark datasets in which the NSS-based GANs outperforms the state-of-the-arts in most cases.