Research map / Image generation / Fast diffusion sampling
A Unified and Fast-Sampling Diffusion Bridge Framework via Stochastic Optimal Control
Mokai Pan, Kaizhen Zhu, Yuexin Ma, Yanwei Fu, Jingyi Yu A, Jingya Wang, Shi Ye
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 Doob's h-transform (infinite penalty) with finite SOC penalty gamma=1e7 (UniDB) (ablation) | GOUB · DIV2K 4x super-resolution | FID: 20.85 -> 16.21 | Improved |
| replace Doob's h-transform with finite penalty gamma=1e7 (UniDB) (ablation) | GOUB · DIV2K 4x super-resolution | PSNR: 26.89 -> 25.46 | Hurt |
| use finite penalty gamma=1e7 (UniDB) (ablation) | GOUB · Raindrop restoration | FID: 36.70 -> 26.07 | Improved |
| increase penalty gamma from 1e7 to 1e8 (ablation) | UniDB · DIV2K 4x super-resolution | FID: 16.21 -> 23.76 | Hurt |
| decrease penalty gamma from 1e7 to 1e6 (ablation) | UniDB · DIV2K 4x super-resolution | PSNR: 25.46 -> 24.72 | Hurt |
| best gamma differs by task: 1e8 on deraining (ablation) | UniDB · Rain100H deraining | PSNR: 32.00 -> 32.05 | Mixed |
| replace Euler sampling of UniDB (NFE 100) with training-free UniDB++ at NFE 5 | UniDB · DIV2K 4x super-resolution | PSNR: 25.46 -> 28.40 | Improved |
| replace Euler sampling of UniDB (NFE 100) with UniDB++ at NFE 5 | UniDB · DIV2K 4x super-resolution | FID: 16.21 -> 17.06 | Hurt |
| increase UniDB++ steps from NFE 5 to 50 | UniDB++ · DIV2K 4x super-resolution | PSNR: 28.40 -> 26.61 | Hurt |
| increase UniDB++ steps from NFE 5 to 50 | UniDB++ · DIV2K 4x super-resolution | FID: 17.06 -> 14.98 | Improved |
| use second-order instead of first-order UniDB++ at large NFE (ablation) | UniDB++ · DIV2K 4x super-resolution | FID: 14.98 -> 14.10 | Improved |
| use second-order instead of first-order UniDB++ at NFE 10 (ablation) | UniDB++ · Raindrop restoration | PSNR: 30.66 -> 29.74 | Hurt |
Closest papers by meaning
- UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control (ICML 2025 · similarity 0.98)
- Exploring the Design Space of Diffusion Bridge Models (NeurIPS 2025 · similarity 0.93)
- Arbitrary-steps Image Super-resolution via Diffusion Inversion (CVPR 2025 · similarity 0.90)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.90)
- Improving Diffusion-Based Image Restoration with Error Contraction and Error Correction (AAAI 2024 · similarity 0.90)
- Image Restoration Through Generalized Ornstein-Uhlenbeck Bridge (ICML 2024 · 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.
