ABCAS: Adaptive Bound Control of spectral norm as Automatic Stabilizer

TL;DR

This paper introduces ABCAS, an adaptive spectral normalization method for GANs that dynamically adjusts the discriminator's Lipschitz constant based on data distribution differences, enhancing training stability and image quality.

Contribution

The paper proposes a novel adaptive normalization technique, ABCAS, which improves GAN training stability by adjusting spectral norm constraints according to data distribution differences.

Findings

Achieved better Fréchet Inception Distance scores.

Demonstrated improved training stability across datasets.

Provided ablation study on spectral norm choices.

Abstract

Spectral Normalization is one of the best methods for stabilizing the training of Generative Adversarial Network. Spectral Normalization limits the gradient of discriminator between the distribution between real data and fake data. However, even with this normalization, GAN's training sometimes fails. In this paper, we reveal that more severe restriction is sometimes needed depending on the training dataset, then we propose a novel stabilizer which offers an adaptive normalization method, called ABCAS. Our method decides discriminator's Lipschitz constant adaptively, by checking the distance of distributions of real and fake data. Our method improves the stability of the training of Generative Adversarial Network and achieved better Fr\'echet Inception Distance score of generated images. We also investigated suitable spectral norm for three datasets. We show the result as an ablation study.