Sparsity Aware Normalization for GANs

TL;DR

This paper introduces sparsity aware normalization (SAN), a novel method for stabilizing GAN training by explicitly considering the sparse feature maps in convolutional networks, leading to improved performance.

Contribution

The paper proposes SAN, a new normalization technique that accounts for sparsity in feature maps, addressing drawbacks of spectral normalization in GANs.

Findings

SAN improves GAN training stability and performance.

SAN requires fewer training epochs and smaller networks.

SAN has negligible computational overhead.

Abstract

Generative adversarial networks (GANs) are known to benefit from regularization or normalization of their critic (discriminator) network during training. In this paper, we analyze the popular spectral normalization scheme, find a significant drawback and introduce sparsity aware normalization (SAN), a new alternative approach for stabilizing GAN training. As opposed to other normalization methods, our approach explicitly accounts for the sparse nature of the feature maps in convolutional networks with ReLU activations. We illustrate the effectiveness of our method through extensive experiments with a variety of network architectures. As we show, sparsity is particularly dominant in critics used for image-to-image translation settings. In these cases our approach improves upon existing methods, in less training epochs and with smaller capacity networks, while requiring practically no computational overhead.