In plain words: Flow matching builds a generator by pushing random noise along smooth paths toward real data, rather than the step-by-step denoising of diffusion models. This guide explains the math from scratch and ships working code with image and text examples so anyone can build their own.
Abstract
Flow Matching (FM) is a recent framework for generative modeling that has achieved state-of-the-art performance across various domains, including image, video, audio, speech, and biological structures. This guide offers a comprehensive and self-contained review of FM, covering its mathematical foundations, design choices, and extensions. By also providing a PyTorch package featuring relevant examples (e.g., image and text generation), this work aims to serve as a resource for both novice and experienced researchers interested in understanding, applying and further developing FM.
Yaron Lipman, Marton Havasi, Peter Holderrieth, Neta Shaul, Matt Le, Brian Karrer, Ricky T. Q. Chen, David Lopez-Paz, Heli Ben-Hamu, Itai Gat
arXiv:2412.06264 · cs.LG · submitted Dec 9, 2024
abstract · pdf