Research map / Image generation / Image restoration
GT-Mean Loss: A Simple Yet Effective Solution for Brightness Mismatch in Low-Light Image Enhancement
Jingxi Liao, Shijie Hao, Richang Hong, Meng Wang
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 GT-mean loss instead of original loss | Uformer · LOLv1 | PSNR: 18.218 -> 18.915 | Improved |
| train with GT-mean loss instead of original loss | CID-Net · LOLv1 | PSNR: 23.809 -> 25.122 | Improved |
| train with GT-mean loss instead of original loss | Uformer · LOLv2-real | PSNR: 14.941 -> 16.103 | Improved |
| train with GT-mean loss instead of original loss | LLFormer · LOLv2-synthetic | PSNR: 24.195 -> 25.152 | Improved |
| train with GT-mean loss instead of original loss | SNR-Aware · LOLv2-synthetic | IAA: 2.209 -> 2.207 | Hurt |
| train with GT-mean loss instead of original loss | CID-Net · LOLv1 | IQA: 4.087 -> 4.074 | Hurt |
| train with GT-mean loss instead of original loss | Restormer · LOLv1 | LPIPS: 0.128 -> 0.122 | Improved |
| train with GT-mean loss instead of original loss | MIRNet · LOLv2-real | LPIPS: 0.313 -> 0.214 | Improved |
| train with GT-mean loss instead of original loss | Five-model average on unpaired sets · DICM/MEF/LIME/NPE/VV (avg) | NIQE: 6.880 -> 6.900 | Mixed |
| replace GT-mean loss with brightness-aligned loss only (ablation) | RetinexFormer · LOLv1 training | PSNR: 18.6 dB to 7.9 dB | Hurt |
| use hybrid schedule switching to brightness-aligned loss at 80K iterations instead of baseline L1 (ablation) | RetinexFormer · LOLv1 training | LPIPS: 0.155 -> 0.129 | Mixed |
| use dynamic weight W (GT-mean loss) instead of fixed phased hybrid (ablation) | RetinexFormer · LOLv1 training | LPIPS, PSNR, brightness discrepancy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- RetinexFormer — Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement (ICCV 2023) · better on 3 of 3 shared results
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 3 of 3 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
- LMT-GP: Combined Latent Mean-Teacher and Gaussian Process for Semi-supervised Low-light Image Enhancement (ECCV 2024 · similarity 0.89)
- Enhancing Low-Light Images: A Synthetic Data Perspective on Practical and Generalizable Solutions (AAAI 2025 · similarity 0.87)
- Zero-Shot Low-Light Image Enhancement via Latent Diffusion Models (AAAI 2025 · similarity 0.87)
- Luminance-Aware Statistical Quantization: Unsupervised Hierarchical Learning for Illumination Enhancement (NeurIPS 2025 · similarity 0.86)
- Interpretable Unsupervised Joint Denoising and Enhancement for Real-World low-light Scenarios (ICLR 2025 · similarity 0.86)
- Improving Visual and Downstream Performance of Low-Light Enhancer with Vision Foundation Models Collaboration (CVPR 2025 · 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.
