license: apache-2.0
OCRFlux-bench-single
OCRFlux-bench-single is a benchmark of 2000 PDF pages and their ground-truth Markdowns sampled from our private document datasets, which are labeled manually with multi-round checking.
This dataset can be used to measure the performance of OCR systems in single-page parsing.
Quick links:
Data Mix
Table 1: Pages breakdown by language
| Language | Pages |
|---|---|
| English | 1000 |
| Chinese | 1000 |
| Total | 2000 |
Data Format
Each row in the dataset corresponds to a single page and its ground-truth Markdown.
Since in our private dataset, all titles and headings of documents are annotated without their heading levels. So in the Markdown, all titles and headings are labeled with H1 level with '# '.
To better represent tables with rowspan or colspan cells, we use HTML format to represent tables instead of the raw Markdown table format in the Markdown strings. Notice that we do not distinguish cells in the table header and table body, which means there are no other tags except <tr> and <td> in the HTML table.
Features:
{
'pdf_name': string, # Name of the PDF document
'language': string, # Language of the PDF document, zh or en
'markdown': string, # Ground-truth Markdown
}
License
This dataset is licensed under Apache-2.0.