Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images
This work provides a significant advancement for researchers and practitioners working on generative models, particularly for image synthesis, by demonstrating a VAE architecture that overcomes previous limitations in both performance and generation speed.
This paper introduces a very deep hierarchical VAE that, for the first time, surpasses the log-likelihood performance of PixelCNN on all natural image benchmarks, while also generating samples thousands of times faster. The VAE achieves higher likelihoods with fewer parameters and is more easily applied to high-resolution images.
We present a hierarchical VAE that, for the first time, generates samples quickly while outperforming the PixelCNN in log-likelihood on all natural image benchmarks. We begin by observing that, in theory, VAEs can actually represent autoregressive models, as well as faster, better models if they exist, when made sufficiently deep. Despite this, autoregressive models have historically outperformed VAEs in log-likelihood. We test if insufficient depth explains why by scaling a VAE to greater stochastic depth than previously explored and evaluating it CIFAR-10, ImageNet, and FFHQ. In comparison to the PixelCNN, these very deep VAEs achieve higher likelihoods, use fewer parameters, generate samples thousands of times faster, and are more easily applied to high-resolution images. Qualitative studies suggest this is because the VAE learns efficient hierarchical visual representations. We release our source code and models at https://github.com/openai/vdvae.