Skip to content

Document parsing and extraction

OmniDocBench

PDF parsing of text, tables, formulas and reading order.

About OmniDocBench

What it measures

Component-specific edit distance, table/formula and ordering metrics.

Scope & limitations

Freeze dataset version, matching rules and render resolution; a single OCR string accuracy hides table and layout failures.

Evaluation environment

Document-to-structured-text/Markdown adapter.

Source

Original benchmark

Dataset & task catalog ↗