High-Fidelity Image Generation With Fewer Labels

TL;DR

This paper introduces a method that leverages self- and semi-supervised learning to generate high-quality images with significantly fewer labeled data, matching or surpassing current state-of-the-art models.

Contribution

The work demonstrates how to reduce label requirements in high-fidelity image generation, achieving comparable or better results than existing models with less labeled data.

Findings

Achieves state-of-the-art FID scores with only 10-20% of labels.

Outperforms BigGAN in image quality using fewer labels.

Effective semi-supervised approach for high-resolution image synthesis.

Abstract

Deep generative models are becoming a cornerstone of modern machine learning. Recent work on conditional generative adversarial networks has shown that learning complex, high-dimensional distributions over natural images is within reach. While the latest models are able to generate high-fidelity, diverse natural images at high resolution, they rely on a vast quantity of labeled data. In this work we demonstrate how one can benefit from recent work on self- and semi-supervised learning to outperform the state of the art on both unsupervised ImageNet synthesis, as well as in the conditional setting. In particular, the proposed approach is able to match the sample quality (as measured by FID) of the current state-of-the-art conditional model BigGAN on ImageNet using only 10% of the labels and outperform it using 20% of the labels.