Research map / Image generation / Image restoration
Bi-Bridge: Bidirectional Diffusion Bridges for Low-Light Image Enhancement
Zeyu Hua, Hui Li, Yu Wang, Song Wang, Congchao Zhu, Caixia Zheng
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 |
|---|---|---|---|
| train with bidirectional (enhancement and degradation) consistency using random start/end swap (ablation) | DDBM · LOL-v2-Real | PSNR: 26.353 -> 31.287 (+4.934 dB) | Improved |
| train with bidirectional consistency (ablation) | DDBM · LOL-v1 | PSNR: 24.451 -> 27.879 | Improved |
| train with bidirectional consistency (ablation) | DDBM · LOL-v2-Synthetic | PSNR: 27.872 -> 31.019 | Improved |
| train with bidirectional consistency (ablation) | DDBM · LOL-v2-Real | LPIPS: 0.165 -> 0.040 | Improved |
| train with bidirectional consistency (ablation) | DDBM · LOL-v2-Real | SSIM: 0.813 -> 0.841 | Improved |
| remove the bidirectional training objective (ablation) | Bi-Bridge · LOL-v1/v2 (NFE 10-80) | PSNR, LPIPS | Hurt |
| use bridge-based formulation without bidirectional objective vs standard DDBM baseline (ablation) | DDBM · LOL-v1/v2 (NFE 10-80) | PSNR, LPIPS | Improved |
| add auxiliary perceptual loss (ablation) | Bi-Bridge · LOL-v1/v2 (NFE 10-80) | PSNR, LPIPS | Improved |
| bidirectional training reduces needed sampling steps (ablation) | Bi-Bridge · LOL-v2-Real | PSNR vs NFE: 20 NFE matches or surpasses 80-NFE baseline (about 4x) | Improved |
| bidirectional bridge instead of general-purpose bidirectional bridge BDBM | Bi-Bridge · LOL-v2-Real | PSNR: 25.102 -> 31.287 | Improved |
| bidirectional training for degradation task vs unidirectional degradation specialist (ablation) | Bi-Bridge · out-of-domain high-resolution images | qualitative degradation fidelity | Improved |
| Bi-Bridge versus top regression and Noise-to-Image diffusion models on SSIM | Bi-Bridge · LOL-v2-Real / LOL-v1 | SSIM: 0.841 vs ReDDiT 0.895; 0.865 vs CIDNet 0.889 | Mixed |
Closest papers by meaning
- IRBridge: Solving Image Restoration Bridge with Pre-trained Generative Diffusion Models (ICML 2025 · similarity 0.90)
- Zero-Shot Low-Light Image Enhancement via Latent Diffusion Models (AAAI 2025 · similarity 0.88)
- Multi-Domain Multi-Scale Diffusion Model for Low-Light Image Enhancement (AAAI 2024 · similarity 0.88)
- AGLLDiff: Guiding Diffusion Models Towards Unsupervised Training-free Real-world Low-light Image Enhancement (AAAI 2025 · similarity 0.88)
- Fourier Priors-Guided Diffusion for Zero-Shot Joint Low-Light Enhancement and Deblurring (CVPR 2024 · similarity 0.87)
- Efficient Diffusion as Low Light Enhancer (CVPR 2025 · 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.
