Research map / Image generation / Editing and style transfer
LightenDiffusion: Unsupervised Low-Light Image Enhancement with Latent-Retinex Diffusion Models
Hai Jiang, Ao Luo, Xiaohong Liu, Songchen Han, Shuaicheng Liu
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 |
|---|---|---|---|
| perform Retinex decomposition in latent space (k=2) instead of image space (k=0) (ablation) | LightenDiffusion · LOL | PSNR: 17.054 -> 20.097 | Improved |
| perform decomposition in latent space speeds inference (ablation) | LightenDiffusion · 400x600 image inference time | Time (s): 4.733 -> 0.411 | Improved |
| increase downsampling scale k from 3 to 4 (ablation) | LightenDiffusion · LOL | PSNR / SSIM: SSIM 0.785 at k=4 | No clear effect |
| use CTDN instead of RetinexNet, URetinexNet or PairLIE decomposition (ablation) | LightenDiffusion · LOL | PSNR: 17.916 (URetinexNet) -> 20.453 | Improved |
| add self-constrained consistency loss L_scc (ablation) | LightenDiffusion · LOL | PSNR: 19.184 -> 20.453 | Improved |
| increase sampling steps to 100 instead of using L_scc (ablation) | LightenDiffusion without L_scc · LOL | PSNR / time: PSNR 20.255, time 1.208 s vs default 20.453 at 0.314 s | Mixed |
| use LightenDiffusion as preprocessing for face detection | RetinaFace on DARK FACE · DARK FACE | AP: 20.2% -> 36.4% | Improved |
| compare to supervised methods on LOL | LightenDiffusion · LOL | PSNR: 20.453 vs SMG 23.814 | Hurt |
Later papers that compared with this method
3 later accepted papers report LightenDiffusion as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- NEC-Diff: Noise-Robust Event-RAW Complementary Diffusion for Seeing Motion in Extreme Darkness (CVPR 2026 · better on 2 of 2 shared results)
- LUCE: Constrained Curve-Domain Guidance for Training-Free Low-Light Enhancement with Hue-Preserving Decoupling (ECCV 2026 · better on 1 of 1 shared result)
- Luminance-Aware Statistical Quantization: Unsupervised Hierarchical Learning for Illumination Enhancement (NeurIPS 2025 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Zero-Shot Low-Light Image Enhancement via Latent Diffusion Models (AAAI 2025 · similarity 0.89)
- AGLLDiff: Guiding Diffusion Models Towards Unsupervised Training-free Real-world Low-light Image Enhancement (AAAI 2025 · similarity 0.89)
- Luminance-Aware Statistical Quantization: Unsupervised Hierarchical Learning for Illumination Enhancement (NeurIPS 2025 · similarity 0.89)
- Interpretable Unsupervised Joint Denoising and Enhancement for Real-World low-light Scenarios (ICLR 2025 · similarity 0.89)
- Zero-Reference Low-Light Enhancement via Physical Quadruple Priors (CVPR 2024 · similarity 0.88)
- IniRetinex: Rethinking Retinex-type Low-Light Image Enhancer via Initialization Perspective (AAAI 2025 · 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.
