Research map / Image generation / Fast diffusion sampling
Diffusion Bridge or Flow Matching? A Unifying Framework and Comparative Analysis
Kaizhen Zhu, Mokai Pan, Zhechuan Yu, Jingya Wang, Jingyi Yu, Ye Shi
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 |
|---|---|---|---|
| use Diffusion Bridge instead of Flow Matching | latent DiT image-to-image · Inpainting Box128 | FID: 17.84 -> 7.71 | Improved |
| use Diffusion Bridge instead of Flow Matching | latent DiT · Nights->Days | FID: 121.53 -> 38.28 | Improved |
| use Diffusion Bridge instead of Flow Matching | latent DiT · Edges->Shoes | FID: 71.38 -> 20.51 | Improved |
| use Diffusion Bridge instead of Flow Matching | latent DiT · Deblurring | SSIM: 0.683 -> 0.661 | Hurt |
| use Diffusion Bridge instead of Flow Matching | latent DiT · Edges->Shoes | MSE: 0.034 -> 0.100 | Hurt |
| add z_T as extra network input to Flow Matching (ablation) | Flow Matching · Inpainting Box128 | FID: 17.84 -> 14.66 | Improved |
| add z_T as extra network input to Flow Matching (ablation) | Flow Matching · Inpainting Box96 | FID: 8.18 -> 7.11 | Improved |
| increase inpainting mask size from 50 to 128 (ablation) | Flow Matching · Inpainting CelebA-HQ | FID: 4.93 -> 17.84 | Hurt |
| increase inpainting mask size from 50 to 128 (ablation) | Diffusion Bridge · Inpainting CelebA-HQ | FID: 4.93 -> 7.71 | Hurt |
| train longer from 40 to 180 epochs (ablation) | Diffusion Bridge · Inpainting Box128 | FID: 25.46 -> 7.71 | Improved |
| train longer from 40 to 180 epochs (ablation) | Flow Matching · Inpainting Box128 | FID: 19.31 -> 18.57 | No clear effect |
| compare DB vs FM at Box50 (ablation) | Diffusion Bridge · Inpainting Box50 | FID: 4.93 vs 4.93 | No clear effect |
Closest papers by meaning
- Inverse Bridge Matching Distillation (ICML 2025 · similarity 0.90)
- UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control (ICML 2025 · similarity 0.90)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.89)
- Exploring the Design Space of Diffusion Bridge Models (NeurIPS 2025 · similarity 0.89)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · similarity 0.89)
- One-step Diffusion with Distribution Matching Distillation (CVPR 2024 · 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.
