# opendatalab/OmniDocBench

[CVPR 2025] A Comprehensive Benchmark for Document Parsing and Evaluation

Repository: https://github.com/opendatalab/OmniDocBench
Canonical: https://ross.abutalabs.com/products/omnidocbench
Homepage: https://opendatalab.com/omnidocbench
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-07-27T09:24:57+00:00

## Health v2 (maintenance only)
Score: 64/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 94, release rhythm 35, longevity 49
- inputs: {"age_days": 688, "days_push": 37, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1999, forks 193 (observed 2026-08-28T04:06:03.624745+00:00)

## What it is
OmniDocBench is a comprehensive benchmark dataset and evaluation toolkit for document parsing, containing 1,651 annotated PDF pages across 10 document types, 5 layouts, and 5 languages. It includes rich block- and span-level annotations (text, tables, formulas, reading order) plus end-to-end and module-level evaluation code with metrics like TEDS, BLEU, and mAP.

## Use cases
- evaluate pdf document parsing models
- benchmark ocr accuracy on real-world documents
- compare table recognition and formula recognition tools
- measure layout detection quality on annotated pdf pages
- test end-to-end document-to-markdown conversion quality

## When to choose
- you need a high-quality annotated dataset to evaluate document parsing systems
- you want standardized metrics (TEDS, BLEU, METEOR, mAP) for comparing parsers
- you need coverage of diverse document types like papers, reports, newspapers, and handwriting

## When to avoid
- you need a production document parser rather than an evaluation benchmark
- your documents fall outside its covered types, layouts, or languages
- you need training data rather than evaluation data

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, ocr, pdf
- domain: computer-vision, pdf, machine-learning
- platform: python, cross-platform
- tags: document-parsing, pdf-parsing, benchmark-dataset, layout-detection, table-recognition, formula-recognition, cvpr-2025, evaluation, natural-language-processing, datasets

## Member repositories
- opendatalab/OmniDocBench (main) score 64

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:03.624745+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:02:10.989538+00:00, confidence not recorded.
  - readme: https://github.com/opendatalab/OmniDocBench (fetched 2026-08-28T04:06:03.624745+00:00, sha e773dc473675)
  - homepage: https://opendatalab.com/omnidocbench (fetched 2026-08-29T10:42:02.154924+00:00, sha 0f12c7c923ab)
- Data as of 2026-08-30T08:39:29.467469+00:00.