Research map / Image generation / Image restoration
Consistency Geodesic Bridge: Image Restoration with Pretrained Diffusion Models
Jinhui HOU, Zhiyu Zhu, Junhui Hou
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 the geodesic trajectory that starts from a degraded-noise mix instead of pure-noise diffusion (ablation) | E-Bridge · Denoising (5 NFE) | LPIPS: 0.316 -> 0.182 | Improved |
| use the geodesic trajectory instead of a conventional bridge with re-noising (ablation) | E-Bridge · Denoising (5 NFE) | LPIPS: 0.235 -> 0.182 | Improved |
| use the geodesic trajectory instead of a conventional bridge (ablation) | E-Bridge · Demoireing (5 NFE) | LPIPS: 0.240 -> 0.230 | Improved |
| use the geodesic trajectory instead of a standard diffusion trajectory (ablation) | E-Bridge · Super-resolution (5 NFE) | LPIPS: 0.495 -> 0.356 | Improved |
| use the consistency solver (10 NFE) instead of 50-step DDIM (ablation) | E-Bridge · Denoising | MUSIQ: 69.01 -> 69.74 | Improved |
| use the consistency solver (10 NFE) instead of 50-step DDIM (ablation) | E-Bridge · Super-resolution | MUSIQ: 69.37 -> 70.87 | Improved |
| use the consistency solver at 1 NFE instead of 1-step DDIM (ablation) | E-Bridge · Super-resolution | PSNR: 17.92 -> 24.09 | Improved |
| increase T0 from 0.3 to 0.9 (ablation) | E-Bridge · Super-resolution | LPIPS: 0.457 -> 0.346 | Improved |
| increase T0 from 0.7 to 0.9 (ablation) | E-Bridge · Denoising | LPIPS: 0.184 -> 0.185 | No clear effect |
| increase T0 from 0.3 to 0.9 (ablation) | E-Bridge · Demoireing | PSNR: 20.85 -> 16.47 | Hurt |
| increase T0 from 0.5 to 0.9 (ablation) | E-Bridge · Raindrop removal | MUSIQ: 66.73 -> 63.98 | Hurt |
| use E-Bridge at 5 NFE instead of UniDB++ at 5 NFE | E-Bridge · Super-resolution | PSNR: 24.393 -> 22.477 | Hurt |
Closest papers by meaning
- UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control (ICML 2025 · similarity 0.88)
- IRBridge: Solving Image Restoration Bridge with Pre-trained Generative Diffusion Models (ICML 2025 · similarity 0.88)
- Image Restoration Through Generalized Ornstein-Uhlenbeck Bridge (ICML 2024 · similarity 0.88)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.87)
- Exploring the Design Space of Diffusion Bridge Models (NeurIPS 2025 · similarity 0.87)
- Consistency Diffusion Bridge Models (NeurIPS 2024 · similarity 0.87)
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.
