Training Flow Matching: The Role of Weighting and Parameterization
This work provides practical insights for researchers and practitioners training flow matching models, but it is incremental as it analyzes existing factors rather than proposing new methods.
The study investigated how loss weighting and output parameterization affect training objectives in denoising-based generative models, finding that these choices interact with data manifold dimensionality, model architecture, and dataset size, with experiments showing impacts on denoising accuracy (e.g., PSNR) and generative quality (e.g., FID).
We study the training objectives of denoising-based generative models, with a particular focus on loss weighting and output parameterization, including noise-, clean image-, and velocity-based formulations. Through a systematic numerical study, we analyze how these training choices interact with the intrinsic dimensionality of the data manifold, model architecture, and dataset size. Our experiments span synthetic datasets with controlled geometry as well as image data, and compare training objectives using quantitative metrics for denoising accuracy (PSNR across noise levels) and generative quality (FID). Rather than proposing a new method, our goal is to disentangle the various factors that matter when training a flow matching model, in order to provide practical insights on design choices.