Overview - ICDAR 2025 Handwritten Notes Understanding Challenge

Handwritten notes contain dense multimodal information, including mathematical equations, diagrams, tables, and domain-specific symbols, which are often lost or misinterpreted by traditional OCR systems.
This ICDAR 2025 Handwritten NoTeS Understanding challenge presents a unique opportunity to push the boundaries of multimodal retrieval, reasoning, and layout-aware question answering.
 

Task 1: Evidence-Based Visual Question Answering (EB-VQA)

Participants must develop models capable of localizing evidence in a handwritten note to answer a given question. Unlike conventional QA, where OCR-extracted text is sufficient, this task requires models to:

• Retrieve key evidence (bounding boxes) from handwritten equations, diagrams, and structured content.
• Generate an answer while preserving the context and meaning of handwritten symbols.
• Provide explicit justification, ensuring explainability in model predictions.

Task 2: Open-Domain Question Answering (OD-QA)

This task tests a model’s ability to search and retrieve information from a large corpus of handwritten notes before answering a question. Participants must:

• Classify the domain (e.g., physics, mathematics) to optimize retrieval efficiency. [optional]
• Retrieve the most relevant document and page, ensuring correct context.
• Generate an answer, with the retrieved evidence (from previous step).