Research map / Image generation / Image restoration
ExposureDiffusion: Learning to Expose for Low-light Image Enhancement
Yufei Wang, Yi Yu, Wenhan Yang, Lanqing Guo, Lap-Pui Chau, Alex C. Kot, Bihan Wen
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 the exposure-diffusion process instead of a feed-forward mapping, same P+G noise model and backbone | UNet · SID Sony x250 | PSNR: 34.39 -> 36.02 | Improved |
| apply ExposureDiffusion to small NAFNet-1 backbone | NAFNet-1 · SID x100 (P+G) | PSNR: 32.37 -> 36.74 | Improved |
| apply ExposureDiffusion on top of SOTA PMN noise model | PMN noise model · SID x100 | PSNR: 39.77 -> 39.80 | No clear effect |
| apply ExposureDiffusion on ELD noise model | UNet · SID x100 | PSNR: 39.27 -> 39.37 | Improved |
| add adaptive residual layer (ablation) | ExposureDiffusion · SID example image | PSNR: 32.57 -> 35.44 | Improved |
| use ExposureDiffusion for unseen amplifying ratios (train x100, test x300) | PMN + NAFNet-2 · SID x300 OOD | PSNR: 36.31 -> 36.83 | Improved |
| increase inference steps to 10 for baseline (ablation) | feed-forward baseline · SID | PSNR | Hurt |
| replace vanilla conditional diffusion with exposure diffusion at similar model size (ablation) | UNet-sized backbone · SID x100 | PSNR: 33.47 -> 38.88 | Improved |
| use L1 loss instead of KL-divergence loss (ablation) | ExposureDiffusion · SID | converged performance | No clear effect |
| small ExposureDiffusion model vs larger feed-forward model | NAFNet-1 vs NAFNet-2 · SID | PSNR and cost: about 25% of parameters and FLOPs | Improved |
| apply ExposureDiffusion on Canon EOS700D x200 vs P+G | UNet · ELD Canon EOS700D x200 | PSNR: 37.86 -> 37.71 | Hurt |
Closest papers by meaning
- Zero-Shot Low-Light Image Enhancement via Latent Diffusion Models (AAAI 2025 · similarity 0.90)
- Multi-Domain Multi-Scale Diffusion Model for Low-Light Image Enhancement (AAAI 2024 · similarity 0.89)
- Efficient Diffusion as Low Light Enhancer (CVPR 2025 · similarity 0.87)
- BlurDM: A Blur Diffusion Model for Image Deblurring (NeurIPS 2025 · similarity 0.87)
- AGLLDiff: Guiding Diffusion Models Towards Unsupervised Training-free Real-world Low-light Image Enhancement (AAAI 2025 · similarity 0.87)
- DPLUT: Unsupervised Low-light Image Enhancement with Lookup Tables and Diffusion Priors (AAAI 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.
