Asymmetric Flow Models
The article discusses a new approach in computer vision called Asymmetric Flow Modeling (AsymFlow). This method improves flow-based generation in high-dimensional spaces by using a rank-asymmetric velocity parameterization. AsymFlow has achieved state-of-the-art results in pixel-space text-to-image generation, outperforming previous models significantly.
- ▪AsymFlow restricts noise prediction to a low-rank subspace while maintaining full-dimensional data prediction.
- ▪On ImageNet 256x256, AsymFlow achieves a leading 1.57 FID, surpassing prior models.
- ▪The model allows for finetuning pretrained latent flow models into pixel-space models, enhancing visual realism.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.12964 (cs) [Submitted on 13 May 2026] Title:Asymmetric Flow Models Authors:Hansheng Chen, Jan Ackermann, Minseo Kim, Gordon Wetzstein, Leonidas Guibas View a PDF of the paper titled Asymmetric Flow Models, by Hansheng Chen and 4 other authors View PDF HTML (experimental) Abstract:Flow-based generation in high-dimensional spaces is difficult because velocity prediction requires modeling high-dimensional noise, even when data has strong low-rank structure. We present Asymmetric Flow Modeling (AsymFlow), a rank-asymmetric velocity parameterization that restricts noise prediction to a low-rank subspace while keeping data prediction full-dimensional.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv.org.
Discussion
0 commentsMore from arXiv.org
