Chunkr Reading Order Bench: 733 Professionally Annotated Documents

733 professionally annotated documents form an open-source benchmark for reading order detection. This dataset contains diverse real-world documents like financial reports, legal contracts, research papers, and medical records, each with ground truth reading order sequences. It was created by ChunkrAI and last updated on 2026-01-22.

Use Cases

• Benchmarking reading order detection models based on the annotated document sequences.
• Training models to understand complex layouts based on the diverse document types mentioned.
• Evaluating model generalization across domains like finance, law, and medicine based on the dataset's composition.

Strengths

• 733 professionally annotated documents provide a substantial testbed.
• Annotations include detailed ground truth reading order sequences.
• Covers diverse, complex real-world document types as described.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

ChunkrAI

Collection Method

Professional annotation of real-world documents.

Freshness

Last updated 2026-01-22 19:53:09.