Quality Aware Generative Adversarial Networks

TL;DR

This paper introduces a novel approach to improve GAN training by incorporating objective image quality metrics, specifically SSIM and NIQE, as regularizers, leading to state-of-the-art results on multiple datasets.

Contribution

It proposes the use of SSIM-based distance metrics and NIQE-inspired gradient penalties as regularizers in GANs, enhancing image quality and training stability.

Findings

Achieved state-of-the-art performance on CIFAR-10, STL10, and CelebA datasets.

Demonstrated the effectiveness of quality-aware regularizers in GAN training.

Improved image quality metrics compared to baseline models.

Abstract

Generative Adversarial Networks (GANs) have become a very popular tool for implicitly learning high-dimensional probability distributions. Several improvements have been made to the original GAN formulation to address some of its shortcomings like mode collapse, convergence issues, entanglement, poor visual quality etc. While a significant effort has been directed towards improving the visual quality of images generated by GANs, it is rather surprising that objective image quality metrics have neither been employed as cost functions nor as regularizers in GAN objective functions. In this work, we show how a distance metric that is a variant of the Structural SIMilarity (SSIM) index (a popular full-reference image quality assessment algorithm), and a novel quality aware discriminator gradient penalty function that is inspired by the Natural Image Quality Evaluator (NIQE, a popular no-reference image quality assessment algorithm) can each be used as excellent regularizers for GAN objective functions. Specifically, we demonstrate state-of-the-art performance using the Wasserstein GAN gradient penalty (WGAN-GP) framework over CIFAR-10, STL10 and CelebA datasets.