Reproduce “Practical Blind Image Denoising via Swin-Conv-UNet and Data Synthesis” 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
Kai Zhang, Yawei Li, Jingyun Liang, Jiezhang Cao, Yulun Zhang, Hao Tang, Deng-Ping Fan, Radu Timofte and 1 more
Swin-Conv-UNet (SCUNet), a hybrid CNN-Transformer architecture, is introduced alongside a comprehensive data synthesis pipeline to enable practical blind image denoising for diverse real-world degradations. This approach achieves state-of-the-art PSNR performance for synthetic Gaussian denoising and demonstrates superior visual quality and robustness on unconstrained real-world noisy images.
Key insights
- A hybrid architecture combining local (CNN) and non-local (Transformer) modeling capabilities within a UNet structure provides a more powerful image prior, leading to enhanced denoising performance.
- Realistic data synthesis, encompassing a wide array of noise types, processing steps (like resizing), and their randomized combinations, is crucial for training deep denoising models that truly generalize to practical blind scenarios.
- Resizing operations significantly alter noise characteristics and must be explicitly accounted for in the data synthesis pipeline to achieve robust practical blind denoising.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.