Skip to content

Dataset · opendatalab

OmniDocBench: Document parsing benchmark

opendatalab/OmniDocBench

A benchmark of 1,651 PDF pages for evaluating document parsing across various types and languages.

What's inside

OmniDocBench is an evaluation dataset for real-world document parsing. It includes 1,651 PDF pages, 10 document types, 5 layout types, and 5 language types. It covers academic papers, reports, newspapers, textbooks, exam papers, magazines, and handwritten and historical documents. Rich annotations are provided.

What you can do with it

  • Evaluating OCR and document parsing
  • Testing vision-language models on documents
  • Comparing layout analysis
Size
1K–10K rows
Downloads this month
46.4K
Likes
124

How to load it

from datasets import load_dataset

ds = load_dataset("opendatalab/OmniDocBench")