Research map / Image generation / Fast diffusion sampling
Riemannian MeanFlow for One-Step Generation on Manifolds
Zichen Zhong, Haoliang Sun, Yukun Zhao, Yongshun Gong, Yilong Yin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add PCGrad conflict-aware multi-task optimization to the decomposed loss (RMF-MT) (ablation) | RMF · Earth Flood | MMD at 1 NFE: 0.068 -> 0.048 | Improved |
| add PCGrad to the decomposed loss (ablation) | RMF · Earth Fire | MMD at 1 NFE: 0.042 -> 0.032 | Improved |
| add PCGrad to the decomposed loss (ablation) | RMF · Earth Volcano (827 samples) | MMD at 1 NFE: 0.092 -> 0.102 | Hurt |
| add PCGrad to the decomposed loss (ablation) | RMF · Torus RNA (7D) | MMD at 1 NFE: 0.20 -> 0.07 | Improved |
| add PCGrad to the decomposed loss (ablation) | RMF · SO(3) datasets | MMD at 1 NFE: Fisher 0.076 -> 0.039 | Improved |
| add PCGrad to the decomposed loss (ablation) | RMF · SE(3) grasping | grasping success rate (%): 65 -> 60 at 1 step | Hurt |
| add PCGrad where gradient cosine similarity is already positive (Glycine) (ablation) | RMF · Torus Glycine | MMD at 1 NFE: 0.04 -> 0.03 | No clear effect |
| add classifier-free guidance to the SO(3) mixture model (ablation) | RMF · SO(3) mixture | MMD to mixture: 0.439 -> 0.234 | Improved |
| use Riemannian MeanFlow instead of Euclidean MeanFlow on high-dimensional hyperspheres (ablation) | RMF · Hypersphere d=128 | MMD at 1 NFE: 0.357 -> 0.026 | Improved |
| use RMF instead of RFM on hyperspheres | RMF · Hypersphere d=3 to 128 | MMD at 1 NFE: d=128: 0.031 -> 0.026 | Improved |
| swap RMF objective for alpha-Flow weighting (ablation) | RMF · Earth spherical datasets | MMD at 1 NFE: Fire 0.042 -> 0.043 | No clear effect |
| swap to improved MeanFlow (iRMF) objective that uses network velocity (ablation) | RMF · Earth spherical datasets | MMD at 1 NFE: Fire 0.042 -> 0.037; Volcano 0.092 -> 0.108 | Mixed |
Closest papers by meaning
- Flow Matching on General Geometries (ICLR 2024 · similarity 0.89)
- Learning Distributions on Manifolds with Free-Form Flows (NeurIPS 2024 · similarity 0.89)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · similarity 0.88)
- Generative Modeling on Manifolds Through Mixture of Riemannian Diffusion Processes (ICML 2024 · similarity 0.88)
- Metric Flow Matching for Smooth Interpolations on the Data Manifold (NeurIPS 2024 · similarity 0.87)
- Categorical Flow Matching on Statistical Manifolds (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
