Normalizing Flows with Iterative Denoising

2026-04-21

AI Summary

This work addresses the performance gap between normalizing flows and diffusion models in high-resolution image generation by proposing iTARFlow, which introduces an iterative denoising mechanism into the normalizing flow framework for the first time. While preserving end-to-end maximum likelihood training and strict invertibility, iTARFlow integrates autoregressive modeling with an iterative sampling strategy inspired by diffusion models, substantially improving generation quality. Experimental results demonstrate that iTARFlow achieves competitive performance on ImageNet at resolutions of 64×64, 128×128, and 256×256, confirming its effectiveness and scalability. The study also uncovers distinctive artifact characteristics inherent to normalizing flows, offering valuable insights for future improvements.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

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.

Problem

• Normalizing Flows
• image generation
• generative modeling
• likelihood-based methods
• iterative denoising

Innovation

• Normalizing Flows
• iterative denoising
• autoregressive generation
• likelihood-based modeling
• iTARFlow