Research map / Image generation / Image restoration
Unsupervised Night Image Enhancement: When Layer Decomposition Meets Light-Effects Suppression
Yeying Jin, Wenhan Yang, Robby T. Tan
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 light-effects suppression network after layer decomposition (ablation) | Decomposition network · real night images | qualitative visibility | Improved |
| add layer decomposition before the suppression network (ablation) | Light-effects suppression network · real night images | qualitative visibility | Improved |
| add light-effects layer G guidance to the suppression network (ablation) | Light-effects suppression network · real night images | qualitative light-effects suppression | Improved |
| add structure and HF-features consistency loss (ablation) | Light-effects suppression network · real night images | qualitative artefacts and details | Improved |
| train suppression and boosting jointly rather than in sequence (ablation) | Suppression then boosting · real night images | qualitative visibility | Improved |
| boost dark regions before suppressing light effects without joint training (ablation) | Sequential pipeline · real night images | qualitative light-effects suppression | Hurt |
| suppress light effects first then boost without joint training (ablation) | Sequential pipeline · real night images | qualitative artefacts | Hurt |
| use proposed unsupervised method over best baseline | Ours vs Yan · GTA5 | PSNR: 26.99 -> 29.79 | Improved |
| use proposed method over best baseline | Ours vs Zhang · Syn-light-effects | PSNR: 16.30 -> 16.95 | Improved |
| use proposed method over KinD++ | Ours vs KinD++ · LOL-test | PSNR: 17.752 -> 21.521 | Improved |
| use proposed method over KinD++ | Ours vs KinD++ · LOL-test | LPIPS: 0.198 -> 0.235 | Hurt |
| use proposed method over DRBN | Ours vs DRBN · LOL-Real | PSNR: 19.66 -> 25.51 | Improved |
Closest papers by meaning
- NightHaze: Nighttime Image Dehazing via Self-Prior Learning (AAAI 2025 · similarity 0.90)
- Interpretable Unsupervised Joint Denoising and Enhancement for Real-World low-light Scenarios (ICLR 2025 · similarity 0.87)
- NightCC: Nighttime Color Constancy via Adaptive Channel Masking (CVPR 2024 · similarity 0.87)
- Unveiling Details in the Dark: Simultaneous Brightening and Zooming for Low-Light Image Enhancement (AAAI 2024 · similarity 0.86)
- VQCNIR: Clearer Night Image Restoration with Vector-Quantized Codebook (AAAI 2024 · similarity 0.86)
- Recasting Regional Lighting for Shadow Removal (AAAI 2024 · similarity 0.85)
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.
