Normalizing Flows with Iterative Denoising

TL;DR

This paper introduces iTARFlow, an improved normalizing flow model that combines autoregressive generation with iterative denoising, achieving competitive image generation performance while maintaining a likelihood-based training objective.

Contribution

The work advances normalizing flow models by integrating iterative denoising with autoregressive sampling, providing a new approach that rivals diffusion models in image synthesis.

Findings

iTARFlow achieves competitive results on ImageNet at multiple resolutions.

The model maintains a fully likelihood-based training process.

Analysis of artifacts offers insights for future improvements.

Abstract

Normalizing Flows (NFs) are a classical family of likelihood-based methods that have received revived attention. Recent efforts such as TARFlow have shown that NFs are capable of achieving promising performance on image modeling tasks, making them viable alternatives to other methods such as diffusion models. In this work, we further advance the state of Normalizing Flow generative models by introducing iterative TARFlow (iTARFlow). Unlike diffusion models, iTARFlow maintains a fully end-to-end, likelihood-based objective during training. During sampling, it performs autoregressive generation followed by an iterative denoising procedure inspired by diffusion-style methods. Through extensive experiments, we show that iTARFlow achieves competitive performance across ImageNet resolutions of 64, 128, and 256 pixels, demonstrating its potential as a strong generative model and advancing the frontier of Normalizing Flows. In addition, we analyze the characteristic artifacts produced by iTARFlow, offering insights that may shed light on future improvements. Code is available at https://github.com/apple/ml-itarflow.