PixelSNAIL: An Improved Autoregressive Generative Model
This improves density estimation for image generation, but it is incremental as it builds on existing causal convolution methods.
The paper tackles the problem of modeling long-range dependencies in autoregressive generative models for high-dimensional data like images, achieving state-of-the-art log-likelihood results of 2.85 bits per dim on CIFAR-10 and 3.80 bits per dim on 32x32 ImageNet.
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