Research map / Image generation / Fast diffusion sampling
Flow Map Learning via Games
Mark Goldstein, Anshuk Uppal, Raghav Singhal, Aahlad Manas Puli, Rajesh Ranganath
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 |
|---|---|---|---|
| replace MeanFlow loss with SGFlow loss | flow map model · CIFAR-10 unconditional (10 sampling steps, 200k training steps) | FID: 37.32 -> 12.26 | Improved |
| replace MeanFlow loss with SGFlow loss | flow map model · CIFAR-10 unconditional (50 sampling steps, 200k training steps) | FID: 4.54 -> 2.88 | Improved |
| replace MeanFlow loss with SGFlow loss | flow map model · CIFAR-10 unconditional (100 sampling steps, 200k training steps) | FID: 4.23 -> 2.81 | Improved |
| replace flow matching with SGFlow | few-step sampling · CIFAR-10 unconditional (10 sampling steps, 200k training steps) | FID: 24.87 -> 12.26 | Improved |
| replace flow matching with SGFlow | many-step sampling · CIFAR-10 unconditional (100 sampling steps, 200k training steps) | FID: 3.05 -> 2.81 | Improved |
| replace Lagrangian map matching loss with SGFlow | flow map model · CIFAR-10 unconditional (10 sampling steps, 200k training steps) | FID: 248.76 -> 12.26 | Improved |
| replace Eulerian map matching loss with SGFlow | flow map model · CIFAR-10 unconditional (100 sampling steps, 200k training steps) | FID: 38.95 -> 2.81 | Improved |
| replace Progressive map matching loss with SGFlow | flow map model · CIFAR-10 unconditional (50 sampling steps, 200k training steps) | FID: 235.20 -> 2.88 | Improved |
| replace MeanFlow with SGFlow | flow map model · CIFAR-10 U-Net backward pass | peak GPU memory (Gb): 14.2 -> 43.2 | Hurt |
| replace Lagrange loss with SGFlow | flow map model · CIFAR-10 U-Net backward pass | peak GPU memory (Gb): 69.8 -> 43.2 | Improved |
Closest papers by meaning
- How to build a consistency model: Learning flow maps via self-distillation (NeurIPS 2025 · similarity 0.90)
- SCoT: Unifying Consistency Models and Rectified Flows via Straight-Consistent Trajectories (NeurIPS 2025 · similarity 0.87)
- Align Your Flow: Scaling Continuous-Time Flow Map Distillation (NeurIPS 2025 · similarity 0.87)
- Improving the Training of Rectified Flows (NeurIPS 2024 · similarity 0.86)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · similarity 0.86)
- Test-time Scaling of Diffusions with Flow Maps (NeurIPS 2026 · similarity 0.86)
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.
