MatchGAN: A Self-Supervised Semi-Supervised Conditional Generative Adversarial Network

TL;DR

MatchGAN introduces a self-supervised semi-supervised approach for conditional GANs that leverages label space augmentation, improving image generation quality with fewer labeled examples on challenging benchmarks.

Contribution

This paper proposes a novel label space-based self-supervised pretext task for semi-supervised conditional GANs, enhancing performance with limited labeled data.

Findings

Outperforms baseline with only 20% labeled data

Achieves better FID, Inception Score, and Attribute Classification Rate

Effective on CelebA and RaFD benchmarks

Abstract

We present a novel self-supervised learning approach for conditional generative adversarial networks (GANs) under a semi-supervised setting. Unlike prior self-supervised approaches which often involve geometric augmentations on the image space such as predicting rotation angles, our pretext task leverages the label space. We perform augmentation by randomly sampling sensible labels from the label space of the few labelled examples available and assigning them as target labels to the abundant unlabelled examples from the same distribution as that of the labelled ones. The images are then translated and grouped into positive and negative pairs by their target labels, acting as training examples for our pretext task which involves optimising an auxiliary match loss on the discriminator's side. We tested our method on two challenging benchmarks, CelebA and RaFD, and evaluated the results using standard metrics including Fr\'{e}chet Inception Distance, Inception Score, and Attribute Classification Rate. Extensive empirical evaluation demonstrates the effectiveness of our proposed method over competitive baselines and existing arts. In particular, our method surpasses the baseline with only 20% of the labelled examples used to train the baseline.