Reproduce “Masked and Shuffled Blind Spot Denoising for Real-World Images” 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
Hamadi Chihaoui, Paolo Favaro
The MASH (Masked and Shuffled Blind Spot Denoising) method introduces an adaptive masking scheme and local pixel shuffling to effectively denoise real-world images, particularly those with spatially correlated noise. It demonstrates competitive, and often superior, performance among single-image self-supervised techniques across datasets like SIDD and FMDD.
Key insights
- The optimal masking ratio in Blind Spot Denoising is critically dependent on noise correlation; higher masking ratios act as a stronger regularizer and are more effective for heavily correlated noise.
- The difference in noise levels estimated by Blind Spot Denoising models with high versus low masking ratios serves as an effective proxy for estimating the spatial correlation of unknown noise.
- Locally permuting pixels within iso-intensity regions of a noisy image can effectively decorrelate noise without distorting the underlying clean signal, thereby improving denoiser robustness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.