Conditional Image Synthesis With Auxiliary Classifier GANs

TL;DR

This paper introduces an improved class-conditional GAN architecture capable of generating high-resolution, globally coherent images at 128x128 resolution, with enhanced discriminability and diversity across ImageNet classes.

Contribution

The paper presents a novel label-conditioned GAN variant that produces high-resolution images and introduces new evaluation methods for assessing sample discriminability and diversity.

Findings

128x128 images are more discriminable than resized 32x32 images across ImageNet classes.

84.7% of classes have diverse samples comparable to real data.

High-resolution samples contain class information not present in low-resolution images.

Abstract

Synthesizing high resolution photorealistic images has been a long-standing challenge in machine learning. In this paper we introduce new methods for the improved training of generative adversarial networks (GANs) for image synthesis. We construct a variant of GANs employing label conditioning that results in 128x128 resolution image samples exhibiting global coherence. We expand on previous work for image quality assessment to provide two new analyses for assessing the discriminability and diversity of samples from class-conditional image synthesis models. These analyses demonstrate that high resolution samples provide class information not present in low resolution samples. Across 1000 ImageNet classes, 128x128 samples are more than twice as discriminable as artificially resized 32x32 samples. In addition, 84.7% of the classes have samples exhibiting diversity comparable to real ImageNet data.