Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images

Abstract

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.

arXiv comment17 pages, 14 figures

References in corpus (16)

• Denoising Diffusion Probabilistic Models
• Gaussian Error Linear Units (GELUs)
• Language Models are Few-Shot Learners
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models
• Conditional Image Generation with PixelCNN Decoders
• NICE: Non-linear Independent Components Estimation
• Density estimation using Real NVP
• Generating Long Sequences with Sparse Transformers
• NVAE: A Deep Hierarchical Variational Autoencoder
• Variational Lossy Autoencoder
• Fixup Initialization: Residual Learning Without Normalization
• Jukebox: A Generative Model for Music
• RNADE: The real-valued neural autoregressive density-estimator
• Diagnosing and Enhancing VAE Models
• Generating High Fidelity Images with Subscale Pixel Networks and Multidimensional Upscaling
• Learnable Explicit Density for Continuous Latent Space and Variational Inference

Cited by in corpus (23)

• Diffusion Models Beat GANs on Image Synthesis
• Improved Denoising Diffusion Probabilistic Models
• Variational Diffusion Models
• VideoGPT: Video Generation using VQ-VAE and Transformers
• Knowledge Distillation in Iterative Generative Models for Improved Sampling Speed
• Learning to Efficiently Sample from Diffusion Probabilistic Models
• Creativity and Machine Learning: A Survey
• Predictive Coding: a Theoretical and Experimental Review
• Self-Supervised Variational Auto-Encoders
• VARA-TTS: Non-Autoregressive Text-to-Speech Synthesis based on Very Deep VAE with Residual Attention
• Clockwork Variational Autoencoders
• Feature Unlearning for Pre-trained GANs and VAEs
• A Contrastive Learning Approach for Training Variational Autoencoder Priors
• Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory
• Re-parameterizing VAEs for stability
• GAN Vocoder: Multi-Resolution Discriminator Is All You Need
• D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
• Relaxed-Responsibility Hierarchical Discrete VAEs
• Diagnosing Vulnerability of Variational Auto-Encoders to Adversarial Attacks
• Learning to infer in recurrent biological networks
• Hierarchical Graph-Convolutional Variational AutoEncoding for Generative Modelling of Human Motion
• Cascading Modular Network (CAM-Net) for Multimodal Image Synthesis
• Boxhead: A Dataset for Learning Hierarchical Representations