Reproduce “High-Resolution Document Shadow Removal via A Large-Scale Real-World Dataset and A Frequency-Aware Shadow Erasing Net” 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

Zinuo Li, Xuhang Chen, Chi-Man Pun, Xiaodong Cun

FSENet uses a Laplacian pyramid to separate broad lighting and color changes from fine text and figure detail, then corrects and recombines them. SD7K contains 7,620 shadow/shadow-free document pairs at mean resolution 2462 × 3699, collected from more than 350 documents. At native resolution, FSENet reaches PSNR 28.67, SSIM 0.96, and RMSE 10.00, versus Mask-ShadowGAN’s 24.67, 0.86, and 15.72, respectively. This comparison does not isolate the dataset and architecture's contributions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.