Reproduce “Uni-DocDiff: A Unified Document Restoration Model Based on Diffusion” 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
Fangmin Zhao, Weichao Zeng, Zhenhang Li, Dongbao Yang, Binbin Li, Xiaojun Bi, Yu Zhou
Researchers at the Institute of Information Engineering, Chinese Academy of Sciences, developed Uni-DocDiff, a unified document restoration model that applies diffusion models within a dual-stream architecture to address various degradations. This model achieves state-of-the-art performance across multiple pixel-level and geometric restoration tasks and demonstrates superior scalability for integrating new functionalities.
Key insights
- Diffusion models are highly effective for document image restoration, particularly in preserving and reconstructing fine-grained details like text edges, overcoming blurriness issues of prior regression-based methods.
- Decoupling geometric transformation tasks (dewarping) into a dedicated, non-iterative branch within a dual-stream architecture significantly enhances the efficiency and robustness of a unified restoration model.
- A strategically designed Prior Pool combined with an adaptive Prior Fusion Module enables effective management of inter-task interference and provides high scalability for seamlessly incorporating new restoration tasks by primarily training the PFM.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.