The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

zontologize data

Test and benchmark data used in zontologize.

tests/

About 1,000 made-up business documents in Dutch and English, used by the tests to check that every supported file kind extracts. Every company, person, project code, number and sentence is invented by a seeded generator; nothing comes from a real organisation.

This folder is not a benchmark. The documents are templated and far from realistic in layout, language and content, so speed or OCR numbers measured on it say little about real folders.

Kind	Files
PDF with a text layer	200
PDF without a text layer, rendered with rotation, blur and noise	120
docx, doc	120, 60
xlsx, xls	80, 40
eml	120
html, txt, csv	80, 80, 60
jpg, png of one page of text	20, 20

tests/manifest.csv lists every file with path, kind, pages, has_text_layer and truth_file.
Each scan and image has a <file>.truth.txt with the text it was rendered from, so OCR output can be checked as a character error rate without hand labelling.

benchmarks/

Build from the original public repository; it is not re-hosted here. It holds about 3,000 real files: Dutch Freedom of Information Act (Woo) documents (DANS, doi:10.17026/dans-zau-e3rk, CC-BY 4.0), Govdocs1 thread 0 (Digital Corpora) and a sample of the Enron Email Dataset (CMU, release 2015-05-07).

Downloads last month: 116

Total file size:

1.43 GB