Skip to content

Dataset · datalab-to

Document and table parsing benchmark

datalab-to/omni_parse_bench

A set of 16,288 tests across 2,937 pages for evaluating the accuracy of text, table, and layout recognition.

What's inside

The benchmark contains document page images and text data to evaluate parsing systems, with a metadata size range of 100k to 1M records. It includes 16,288 test tasks across 2,937 pages, used to compare the extraction quality of text, tables, and layout structure.

What you can do with it

  • Evaluating the quality of multimodal models and OCR systems when parsing complex documents
  • Comparing table extraction accuracy and layout structure preservation across different parsers
Tasks
Image → text
Size
100K–1M rows
Downloads this month
7.1K
Likes
22

How to load it

from datasets import load_dataset

ds = load_dataset("datalab-to/omni_parse_bench")