SFBD Flow: A Continuous-Optimization Framework for Training Diffusion Models with Noisy Samples

TL;DR

SFBD flow is a continuous-optimization framework that improves training of diffusion models on noisy data, reducing manual effort and outperforming baselines.

Contribution

It reinterprets SFBD as an alternating projection, introduces a continuous variant called SFBD flow, and demonstrates its effectiveness with an online implementation.

Findings

Online SFBD outperforms strong baselines across benchmarks.

SFBD flow removes manual coordination in training diffusion models.

Connection established between SFBD flow and consistency constraint methods.

Abstract

Diffusion models achieve strong generative performance but often rely on large datasets that may include sensitive content. This challenge is compounded by the models' tendency to memorize training data, raising privacy concerns. SFBD (Lu et al., 2025) addresses this by training on corrupted data and using limited clean samples to capture local structure and improve convergence. However, its iterative denoising and fine-tuning loop requires manual coordination, making it burdensome to implement. We reinterpret SFBD as an alternating projection algorithm and introduce a continuous variant, SFBD flow, that removes the need for alternating steps. We further show its connection to consistency constraint-based methods, and demonstrate that its practical instantiation, Online SFBD, consistently outperforms strong baselines across benchmarks.