Research map / Image generation / Fast diffusion sampling
Direct Diffusion Bridge using Data Consistency for Inverse Problems
Hyungjin Chung, Jeongsol Kim, Jong Chul 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 |
|---|---|---|---|
| Add shallow data-consistency gradient steps | I²SB sampling · ImageNet SR×4-bicubic | PSNR: 25.22 -> 26.41 | Improved |
| Add shallow data-consistency gradient steps | I²SB sampling · ImageNet SR×4-bicubic | FID: 24.13 -> 19.88 | Improved |
| Add shallow data-consistency gradient steps | I²SB sampling · ImageNet deblur-gauss | PSNR: 36.01 -> 37.02 | Improved |
| Add shallow data-consistency gradient steps | I²SB sampling · ImageNet deblur-uniform | PSNR: 30.75 -> 31.26 | Improved |
| Use data-consistency gradient steps for JPEG restoration | I²SB · ImageNet JPEG restoration QF=10 | PSNR: 26.12 -> 26.34 | Improved |
| Set the gradient step-size coefficient to 0.0 instead of 1.0 (ablation) | CDDB · ImageNet SR×4-bicubic, 100 images | PSNR: 26.31 -> 25.07 | Hurt |
| Increase CDDB step-size coefficient from 1.0 to 3.0 (ablation) | CDDB · ImageNet SR×4-bicubic, 100 images | PSNR: 26.31 -> 25.70 | Hurt |
| Use deeper gradients instead of shallow gradients at c=1.0 (ablation) | CDDB · ImageNet SR×4-bicubic, 100 images | PSNR: 26.31 -> 26.56 | Improved |
| Increase CDDB-deep step-size coefficient from 1.0 to 1.5 (ablation) | CDDB-deep · ImageNet SR×4-bicubic, 100 images | PSNR: 26.56 -> 26.59 | Improved |
| Increase CDDB-deep step-size coefficient from 1.0 to 3.0 (ablation) | CDDB-deep · ImageNet SR×4-bicubic, 100 images | PSNR: 26.56 -> 26.06 | Hurt |
| Apply multiple gradient-descent steps to intermediate denoised estimates (ablation) | CDDB · ImageNet SR×4-bicubic | PSNR | Improved |
| Use shallow consistency gradients rather than deeper U-Net-Jacobian gradients (ablation) | CDDB · Inpainting | quantitative metrics | No clear effect |
Closest papers by meaning
- SITCOM: Step-wise Triple-Consistent Diffusion Sampling For Inverse Problems (ICML 2025 · similarity 0.91)
- DMPlug: A Plug-in Method for Solving Inverse Problems with Diffusion Models (NeurIPS 2024 · similarity 0.90)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.90)
- Diffusion Bridge Implicit Models (ICLR 2025 · similarity 0.90)
- Self-diffusion for Solving Inverse Problems (NeurIPS 2025 · similarity 0.89)
- Fast samplers for Inverse Problems in Iterative Refinement models (NeurIPS 2024 · similarity 0.89)
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.
