Reproduce “AIM 2025 Challenge on Real-World RAW Image Denoising” 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
Feiran Li, Jiacheng Li, Marcos V. Conde, Beril Besbinar, Vlad Hosu, Daisuke Iso, Radu Timofte
This report details the "AIM 2025 Challenge on Real-World RAW Image Denoising," which aimed to advance camera-agnostic denoising solutions trained using data synthesis. The competition established a new benchmark and identified top-performing methods, with the MR-CAS team achieving the highest overall ranking with 41.90 PSNR and 0.9633 SSIM.
Key insights
- Data synthesis plays a crucial role in training robust, generalizable denoising models, with top teams employing meticulous noise modeling and additional datasets.
- Effective camera-agnostic denoising benefits from specialized architectural components (e.g., frequency modulation, random masking, group convolutions, attention mechanisms) tailored for RAW image characteristics.
- Progressive multi-stage training strategies and hybrid loss functions, often built upon powerful backbones like NAFNet or Restormer, consistently yielded high performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.