Hierarchical Autoregressive Image Models with Auxiliary Decoders

TL;DR

This paper introduces hierarchical autoregressive image models with auxiliary decoders that learn abstract representations, enabling the generation of high-resolution, large-scale coherent images and outperforming existing models in realism.

Contribution

It proposes a novel hierarchical approach with auxiliary decoders to improve large-scale coherence in autoregressive image generation.

Findings

Models generate realistic 128x128 and 256x256 images.

Hierarchical models outperform non-hierarchical counterparts.

Human evaluation favors the proposed models over state-of-the-art alternatives.

Abstract

Autoregressive generative models of images tend to be biased towards capturing local structure, and as a result they often produce samples which are lacking in terms of large-scale coherence. To address this, we propose two methods to learn discrete representations of images which abstract away local detail. We show that autoregressive models conditioned on these representations can produce high-fidelity reconstructions of images, and that we can train autoregressive priors on these representations that produce samples with large-scale coherence. We can recursively apply the learning procedure, yielding a hierarchy of progressively more abstract image representations. We train hierarchical class-conditional autoregressive models on the ImageNet dataset and demonstrate that they are able to generate realistic images at resolutions of 128$\times$128 and 256$\times$256 pixels. We also perform a human evaluation study comparing our models with both adversarial and likelihood-based state-of-the-art generative models.