PixelSNAIL: An Improved Autoregressive Generative Model

TL;DR

PixelSNAIL introduces a novel autoregressive generative model combining causal convolutions with self-attention, achieving state-of-the-art density estimation results on image datasets like CIFAR-10 and ImageNet.

Contribution

The paper presents a new architecture that enhances autoregressive models with self-attention, improving long-range dependency modeling in image density estimation.

Findings

Achieved 2.85 bits per dim on CIFAR-10

Achieved 3.80 bits per dim on 32x32 ImageNet

Outperformed previous state-of-the-art models

Abstract

Autoregressive generative models consistently achieve the best results in density estimation tasks involving high dimensional data, such as images or audio. They pose density estimation as a sequence modeling task, where a recurrent neural network (RNN) models the conditional distribution over the next element conditioned on all previous elements. In this paradigm, the bottleneck is the extent to which the RNN can model long-range dependencies, and the most successful approaches rely on causal convolutions, which offer better access to earlier parts of the sequence than conventional RNNs. Taking inspiration from recent work in meta reinforcement learning, where dealing with long-range dependencies is also essential, we introduce a new generative model architecture that combines causal convolutions with self attention. In this note, we describe the resulting model and present state-of-the-art log-likelihood results on CIFAR-10 (2.85 bits per dim) and $32 \times 32$ ImageNet (3.80 bits per dim). Our implementation is available at https://github.com/neocxi/pixelsnail-public