Benchmark RadarMPDocBench-Parse
MPDocBench-Parse is a benchmark for multi-page document parsing, covering 433 manually annotated documents with 3,246 pages across 15 document types in English and Chinese. It evaluates content fidelity and logical structure, including text, table, formula recognition, reading order, and heading hierarchy recovery.
- Released
- 2026-05-21
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
The benchmark addresses the lack of realistic multi-page document parsing evaluation, offering a more comprehensive protocol for assessing semantic continuity, hierarchical structure recovery, and visual content preservation, which is crucial for advancing document parsing systems in practical applications.
Motivation
Document parsing converts visually rich documents into machine-readable structured representations, forming a crucial foundation for information systems.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.