Reproduce “FusionNet: Multi-model Linear Fusion Framework for Low-light Image Enhancement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kangbiao Shi, Yixu Feng, Tao Hu, Yu Cao, Peng Wu, Yijin Liang, Yanning Zhang, Qingsen Yan
FusionNet presents a linear fusion framework that integrates multiple independently trained low-light image enhancement models, securing first place in the CVPR 2025 NTIRE Low Light Enhancement Challenge. This approach improves image visibility, color fidelity, and quality by combining diverse architectural and color space strengths while avoiding common fusion strategy drawbacks.
Key insights
- A simple linear fusion of independently trained models can achieve superior low-light enhancement performance and stability compared to more complex serial or multi-stage fusion architectures.
- Different model architectures and color spaces offer complementary strengths in addressing various aspects of image degradation, which can be effectively combined.
- The parallel execution and linear combination strategy helps mitigate common issues like parameter explosion, optimization instability, and feature misalignment in fusion-based approaches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.