Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Straight Flows: Variational Flow Matching for Efficient Generation
Chenrui Ma, Xi Xiao, Tianyang Wang, Xiao Wang, Yanning Shen
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 variational latent code and straightness objective to flow matching (S-VFM vs Flow Matching) | Flow Matching on CIFAR-10 · CIFAR-10, NFE=10 | FID: 14.4 -> 1.97 | Improved |
| add straightness objective on top of VFM (S-VFM vs VFM) (ablation) | VFM on CIFAR-10 · CIFAR-10, NFE=10 | FID: 5.34 -> 1.97 | Improved |
| add straightness objective on top of VFM (S-VFM vs VFM) (ablation) | VFM on CIFAR-10 · CIFAR-10, NFE=2 | FID: 97.83 -> 2.16 | Improved |
| condition on latent via adaptive normalization instead of bottleneck sum (ablation) | S-VFM on CIFAR-10 · CIFAR-10, NFE=1 | FID: 2.94 -> 2.81 | Improved |
| increase NFE for S-VFM (ablation) | S-VFM on CIFAR-10 · CIFAR-10 | FID: 2.81 -> 1.97 | Improved |
| increase NFE for MeanFlow | MeanFlow on CIFAR-10 · CIFAR-10 | FID: 2.23 -> 2.84 | Hurt |
| S-VFM-XL instead of MeanFlow-XL at one step | SiT-XL/2 on ImageNet 256x256 · ImageNet 256x256, NFE=1 | FID: 3.43 -> 3.31 | Improved |
| S-VFM-XL instead of MeanFlow-XL at two steps | SiT-XL/2 on ImageNet 256x256 · ImageNet 256x256, NFE=2 | FID: 2.93 -> 2.86 | Improved |
| S-VFM adaptive norm vs IMM at two steps on CIFAR-10 | CIFAR-10 · CIFAR-10, NFE=2 | FID: 1.98 -> 2.16 | Hurt |
Closest papers by meaning
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.92)
- Variational Flow Matching for Graph Generation (NeurIPS 2024 · similarity 0.90)
- Variational Rectified Flow Matching (ICML 2025 · similarity 0.89)
- Structured Coupling for Flow Matching (NeurIPS 2026 · similarity 0.89)
- On the Guidance of Flow Matching (ICML 2025 · similarity 0.88)
- Beyond Trajectory Matching: Reflow with Marginal Distribution Alignment (NeurIPS 2026 · similarity 0.88)
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.
