A Unified Generative Adversarial Network Training via Self-Labeling and Self-Attention

TL;DR

This paper introduces a unified GAN training method that uses self-labeling and self-attention to improve data quality, effectively handling various levels of labeling and outperforming some existing models.

Contribution

The authors propose a novel GAN training scheme combining artificial labeling, self-labeling, and self-attention, which enhances data quality and can outperform class-conditional GANs.

Findings

Self-labeling and self-attention improve generated data quality.

The method outperforms class-conditional GANs on benchmark datasets.

The approach effectively handles different levels of labeling.

Abstract

We propose a novel GAN training scheme that can handle any level of labeling in a unified manner. Our scheme introduces a form of artificial labeling that can incorporate manually defined labels, when available, and induce an alignment between them. To define the artificial labels, we exploit the assumption that neural network generators can be trained more easily to map nearby latent vectors to data with semantic similarities, than across separate categories. We use generated data samples and their corresponding artificial conditioning labels to train a classifier. The classifier is then used to self-label real data. To boost the accuracy of the self-labeling, we also use the exponential moving average of the classifier. However, because the classifier might still make mistakes, especially at the beginning of the training, we also refine the labels through self-attention, by using the labeling of real data samples only when the classifier outputs a high classification probability score. We evaluate our approach on CIFAR-10, STL-10 and SVHN, and show that both self-labeling and self-attention consistently improve the quality of generated data. More surprisingly, we find that the proposed scheme can even outperform class-conditional GANs.