Flow++: Improving Flow-Based Generative Models with Variational Dequantization and Architecture Design

TL;DR

Flow++ introduces enhancements to flow-based generative models through variational dequantization and improved architecture, significantly boosting their density estimation performance to rival autoregressive models on image benchmarks.

Contribution

The paper proposes Flow++, a novel flow-based model with variational dequantization and advanced architecture, achieving state-of-the-art results in density estimation.

Findings

Flow++ outperforms previous flow-based models on standard image benchmarks.

Flow++ narrows the performance gap between flow-based and autoregressive models.

The implementation is publicly available for reproducibility.

Abstract

Flow-based generative models are powerful exact likelihood models with efficient sampling and inference. Despite their computational efficiency, flow-based models generally have much worse density modeling performance compared to state-of-the-art autoregressive models. In this paper, we investigate and improve upon three limiting design choices employed by flow-based models in prior work: the use of uniform noise for dequantization, the use of inexpressive affine flows, and the use of purely convolutional conditioning networks in coupling layers. Based on our findings, we propose Flow++, a new flow-based model that is now the state-of-the-art non-autoregressive model for unconditional density estimation on standard image benchmarks. Our work has begun to close the significant performance gap that has so far existed between autoregressive models and flow-based models. Our implementation is available at https://github.com/aravindsrinivas/flowpp