HANS: A Handwritten Answer Sheet Dataset for Noisy Hybrid Document Parsing
Researchers created HANS, a new dataset of 5,213 handwritten answer sheets from eighth-grade math exams in China, featuring complex interleaving of text and math and various noise like strikethroughs. They then developed NA-GOT, an end-to-end recognition framework that uses a two-stage noise suppression system—a feature-level module and a noise-aware attention mechanism—to handle this heavy noise effectively. The results show NA-GOT significantly improves accuracy and stability compared to existing methods.