Research map / Probabilistic and scientific modeling / Variational and generative sampling
Contrastive Flow Matching
George Stoica, Vivek Ramanujan, Xiang Fan, Ali Farhadi, Ranjay Krishna, Judy Hoffman
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 contrastive flow term (DeltaFM) | SiT-B/2 · ImageNet-1k 256x256 | FID: 42.28 -> 33.39 | Improved |
| add contrastive flow term (DeltaFM) | SiT-XL/2 · ImageNet-1k 256x256 | FID: 20.01 -> 16.32 | Improved |
| add contrastive flow term (DeltaFM) | SiT-XL/2 · ImageNet-1k 512x512 | FID: 22.98 -> 19.67 | Improved |
| add contrastive flow term (DeltaFM) | SiT-XL/2 · ImageNet-1k 256x256 | recall: 0.63 -> 0.63 | No clear effect |
| add contrastive flow term on top of REPA | REPA SiT-XL/2 · ImageNet-1k 256x256 | FID: 11.14 -> 7.29 | Improved |
| add contrastive flow term on top of REPA | REPA SiT-B/2 · ImageNet-1k 256x256 | FID: 27.33 -> 20.52 | Improved |
| add contrastive flow term on top of REPA | REPA-MMDiT · CC3M 256x256 | FID: 24 -> 19 | Improved |
| add contrastive flow term with adapted CFG | REPA SiT-XL/2 · ImageNet-1k 256x256 with CFG | FID: 2.09 -> 1.97 | Improved |
| increase lambda from 0.0 to 0.05 (ablation) | DeltaFM REPA SiT-XL/2 · ImageNet-1k 256x256 | FID: 11.14 -> 7.29 | Improved |
| increase lambda too high to 0.15 (ablation) | DeltaFM REPA SiT-XL/2 · ImageNet-1k 256x256 | FID: 7.29 -> 19.21 | Hurt |
| use very small lambda 0.001 (ablation) | DeltaFM REPA SiT-XL/2 · ImageNet-1k 256x256 | FID: 11.14 -> 10.93 | No clear effect |
| increase batch size from 256 to 512 with DeltaFM | REPA SiT-XL/2 · ImageNet-1k 256x256 | FID: 7.29 -> 6.36 | Improved |
Closest papers by meaning
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · similarity 0.90)
- Discrete Flow Matching (NeurIPS 2024 · similarity 0.90)
- Value Gradient Guidance for Flow Matching Alignment (NeurIPS 2025 · similarity 0.90)
- On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity (NeurIPS 2025 · similarity 0.90)
- CAR-Flow: Condition-Aware Reparameterization Aligns Source and Target for Better Flow Matching (NeurIPS 2025 · similarity 0.90)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.90)
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.
