Research map / Image generation / Fast diffusion sampling
Training-Free Refinement of Flow Matching with Divergence-based Sampling
Yeonwoo Cha, Jaehoon Yoo, Semin Kim, Yunseo Park, Jinhyeon Kwon, Seunghoon Hong
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 FDS to Euler solver | EDM · CIFAR-10 (Euler 50 NFE) | FID: 3.04 -> 2.44 | Improved |
| add FDS to Euler solver at 100 steps | EDM Cond. · CIFAR-10 | FID: 2.352 -> 1.907 | Improved |
| add FDS to Euler solver at 100 steps | JiT-B/16 · ImageNet 256 | FID: 3.859 -> 3.579 | Improved |
| add FDS to Heun solver at 199 NFE | JiT-L/16 · ImageNet 256 | FID: 2.811 -> 2.579 | Improved |
| add FDS versus giving the same compute to more Euler steps | JiT-B/16 · ImageNet 256 | FID: 3.857 -> 3.579 | Improved |
| add FDS versus giving the same compute to more Heun steps | EDM Cond. · CIFAR-10 | FID: 1.924 -> 1.785 | Improved |
| add FDS to Euler solver at 50 steps | SiT-XL · ImageNet 256 | FID: 2.417 -> 2.291 | Improved |
| add FDS to Heun solver at 99 NFE | SiT-XL · ImageNet 256 | FID: 2.153 -> 2.103 | Improved |
| add FDS to SD3.5-M | SD3.5-M · DrawBench | ImageReward: 81.05 -> 83.60 | Improved |
| add FDS to SD3-M at CFG 7.0 | SD3-M · MS-COCO zero-shot | FID: 23.11 -> 22.18 | Improved |
| add FDS to SD3-M at CFG 3.0 | SD3-M · MS-COCO zero-shot | CLIP score: 25.96 -> 26.02 | No clear effect |
| apply FDS only at early timesteps up to T_trunc (ablation) | FDS on JiT-B/16 · ImageNet 256 | FID: saturates around T_trunc=0.5 | Improved |
Closest papers by meaning
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.89)
- Simple ReFlow: Improved Techniques for Fast Flow Models (ICLR 2025 · similarity 0.89)
- Coreset-Induced Conditional Velocity Flow Matching (NeurIPS 2026 · similarity 0.89)
- Reflected Flow Matching (ICML 2024 · similarity 0.89)
- Elucidating Flow Matching ODE Dynamics via Data Geometry and Denoisers (ICML 2025 · similarity 0.89)
- On Fitting Flow Models with Large Sinkhorn Couplings (NeurIPS 2026 · similarity 0.89)
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.
