Research map / Architectures and efficiency / Mixture of experts
GM-MoE: Low-Light Enhancement with Gated-Mechanism Mixture-of-Experts
Minwen Liao, Haobo Dong, Xinyi Wang, Kurban Ubul, Yihua Shao, Ziyang Yan
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 the shallow feature extraction block (SFEB) to the baseline (ablation) | GM-MoE baseline · LOLv2-Synthetic | PSNR: PSNR improved by 3.09 dB and SSIM improved by 0.0215 | Improved |
| add Expert1 (color restoration) to baseline+SFEB (ablation) | GM-MoE · LOLv2-Real | PSNR: PSNR increased by 1.08 dB (20.27 -> 21.35) | Improved |
| add Expert2 (detail enhancement) after Expert1 (ablation) | GM-MoE · LOLv2-Real and LOLv2-Synthetic | PSNR: PSNR increased by 0.76 dB (LOL-v2-real) and 0.79 dB (LOL-v2-syn), SSIM increased by 0.0575 and 0.0891 | Improved |
| add the gated weight generation network to the three-expert model without Net3 (ablation) | GM-MoE · LOLv2-Real | PSNR: 22.11 -> 23.23 | Improved |
| add Expert3 to the model with gating and Experts 1-2 (ablation) | GM-MoE · LOLv2-Real | PSNR: 23.23 -> 23.65 | Improved |
| add Expert2 to the model with gating and Experts 1 and 3 (ablation) | GM-MoE · LOLv2-Real | PSNR: 23.31 -> 23.65 | Improved |
| add Expert1 to the model with gating and Experts 2 and 3 (ablation) | GM-MoE · LOLv2-Real | PSNR: 23.35 -> 23.65 | Improved |
| use dynamic gating instead of fixed expert weights (ablation) | GM-MoE · LOLv2-Real | PSNR: 21.70 -> 23.65 | Improved |
| mix experts in parallel with gating instead of serializing them (ablation) | GM-MoE · LOLv2-Real | PSNR: PSNR 21.34 -> 23.65, SSIM 0.83 -> 0.80 | Mixed |
| use gated weighted sum instead of channel concat plus 1x1 fusion of experts (ablation) | GM-MoE · LOLv2-Real | PSNR: 17.84 -> 23.65 | Improved |
| GM-MoE versus the next best method on LSRW datasets | GM-MoE vs Restormer · LSRW-Huawei and LSRW-Nikon | PSNR: PSNR improvements of 0.94 dB and 1.42 dB | Improved |
Closest papers by meaning
- Empowering Low-Light Image Enhancer through Customized Learnable Priors (ICCV 2023 · similarity 0.86)
- Low-Light Image Enhancement with Normalizing Flow (AAAI 2022 · similarity 0.85)
- Low-Light Image Enhancement with Illumination-Aware Gamma Correction and Complete Image Modelling Network (ICCV 2023 · similarity 0.85)
- Low-Light Image Enhancement Network Based on Multi-Scale Feature Complementation (AAAI 2023 · similarity 0.85)
- Enhancing Low-Light Images: A Synthetic Data Perspective on Practical and Generalizable Solutions (AAAI 2025 · similarity 0.84)
- Improving Visual and Downstream Performance of Low-Light Enhancer with Vision Foundation Models Collaboration (CVPR 2025 · similarity 0.84)
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.
