Datasets:

ocr-research
/

messy-ocr-content

This dataset contains 100 unique publicly-accessible pdf documents from the internet, intended for Optical Character Recognition (OCR) tasks. The team currently uses this data for vLLM preliminary testing.

Dataset Details

The 100 documents are divided equally into four content groups :

Folder 1, Clear Tables : documents with clear, structured, and outlined tables. Handwritten text included.
Folder 2, Unclear Tables : documents with unclear, multi-column texts with no outlines. Handwritten text included.
Folder 3, Charts & Figures : documents with many charts, figures, seals, or signatures.
Folder 4, Formulas : documents or snippets of published articles containing formulas.

Personal and Sensitive Information

If you are an owner of any file in this dataset and would like to request data take down, contact us as soon as possible.