Reproduce “YOND: Practical Blind Raw Image Denoising Free from Camera-Specific Data Dependency” 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
Hansen Feng, Lizhi Wang, Yiqi Huang, Tong Li, Lin Zhu, Hua Huang
YOND introduces a practical blind raw image denoising method that eliminates camera-specific data dependency by enhancing traditional noise estimation and variance stabilization with learning-based components. The approach achieves state-of-the-art performance across diverse unknown cameras and datasets, providing robust generalization and controllable denoising capabilities.
Key insights
- Using a coarse denoised image to refine noise estimation significantly improves the accuracy and stability of noise parameter estimation across diverse camera and scene conditions.
- Correcting VST expectation bias based on the noisy image *before* denoising prevents amplification of errors and preserves image priors, crucial for accurate color rendition, especially in low-light.
- Integrating an explicit SNR-guided branch into the denoiser allows for controllable denoising strength, enabling adaptive adjustments and manual fine-tuning for diverse real-world scenarios and user preferences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.