StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasets

TL;DR

This paper introduces StyleGAN-XL, a scalable version of StyleGAN that leverages a new training strategy to generate high-quality, diverse images at large dataset scales like ImageNet, surpassing previous limitations.

Contribution

The paper presents StyleGAN-XL, a novel training approach enabling StyleGAN3 to effectively learn from large, unstructured datasets such as ImageNet, achieving state-of-the-art results.

Findings

Successfully trained StyleGAN3 on ImageNet at 1024x1024 resolution.

Achieved state-of-the-art image synthesis quality on large-scale datasets.

Demonstrated image inversion and editing beyond portrait and object-specific domains.

Abstract

Computer graphics has experienced a recent surge of data-centric approaches for photorealistic and controllable content creation. StyleGAN in particular sets new standards for generative modeling regarding image quality and controllability. However, StyleGAN's performance severely degrades on large unstructured datasets such as ImageNet. StyleGAN was designed for controllability; hence, prior works suspect its restrictive design to be unsuitable for diverse datasets. In contrast, we find the main limiting factor to be the current training strategy. Following the recently introduced Projected GAN paradigm, we leverage powerful neural network priors and a progressive growing strategy to successfully train the latest StyleGAN3 generator on ImageNet. Our final model, StyleGAN-XL, sets a new state-of-the-art on large-scale image synthesis and is the first to generate images at a resolution of $1024^2$ at such a dataset scale. We demonstrate that this model can invert and edit images beyond the narrow domain of portraits or specific object classes.