Does Diffusion Beat GAN in Image Super Resolution?

TL;DR

This paper rigorously compares diffusion-based and GAN-based image super resolution models under controlled conditions, revealing that GANs can match or outperform diffusion models when scaled equally, and examines the effects of design choices.

Contribution

It provides a fair comparison between diffusion and GAN models for ISR, showing GANs can be competitive, and analyzes how design choices influence performance.

Findings

GAN models can achieve comparable or better results than diffusion models when scaled equally.

Design choices like text conditioning and augmentation significantly impact ISR performance.

Controlled experiments reveal the true capabilities of each approach independent of model size.

Abstract

There is a prevalent opinion that diffusion-based models outperform GAN-based counterparts in the Image Super Resolution (ISR) problem. However, in most studies, diffusion-based ISR models employ larger networks and are trained longer than the GAN baselines. This raises the question of whether the high performance stems from the superiority of the diffusion paradigm or if it is a consequence of the increased scale and the greater computational resources of the contemporary studies. In our work, we thoroughly compare diffusion-based and GAN-based Super Resolution models under controlled settings, with both approaches having matched architecture, model and dataset sizes, and computational budget. We show that a GAN-based model can achieve results comparable or superior to a diffusion-based model. Additionally, we explore the impact of popular design choices, such as text conditioning and augmentation on the performance of ISR models, showcasing their effect in several downstream tasks. We will release the inference code and weights of our scaled GAN.