Training Flow Matching: The Role of Weighting and Parameterization
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908972733169664 |
|---|---|
| author | Gagneux, Anne Martin, Ségolène Gribonval, Rémi Massias, Mathurin |
| author_facet | Gagneux, Anne Martin, Ségolène Gribonval, Rémi Massias, Mathurin |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_06454 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Training Flow Matching: The Role of Weighting and Parameterization Gagneux, Anne Martin, Ségolène Gribonval, Rémi Massias, Mathurin Computer Vision and Pattern Recognition 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. |
| title | Training Flow Matching: The Role of Weighting and Parameterization |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.06454 |