Open Research Data · Release 2026-09
OCR benchmark dataset
This is the September 2026 fixture set for the in-browser Tesseract run: 10 controlled document scenes, with character and word error rates. It is not an industry benchmark and not a real-user sample. Download the JSON, the CSV, and the fixture manifest.
OCR benchmark dataset?
This is the September 2026 fixture set for the in-browser Tesseract run: 10 controlled document scenes, with character and word error rates. It is not an industry benchmark and not a real-user sample. Download the JSON, the CSV, and the fixture manifest.
Last updated
Downloadable Artifacts
All numbers published in this benchmark are derived from these machine-readable artifacts.
results.json
Full benchmark output including engine and E2E summaries, per-language metrics, timing, and error flags.
results.csv
Flat tabular format with columns for language, pack, CER, WER, key fields, numeric accuracy, and latency.
fixtures.json
Fixture manifest containing public IDs, ground truth strings, verification dates, and scenario metadata.
Dataset Schema & Field Reference
| Field | Type | Description |
|---|---|---|
| public_id | string | Stable public identifier (e.g. OCR-EN-2026-001 to OCR-AR-2026-010). |
| pack | string | Tesseract language model code (e.g. eng, deu, chi_sim, jpn). |
| cer | float | Character Error Rate (Levenshtein distance / total reference characters). Lower is better. |
| wer | float | null | Word Error Rate on space-delimited words. Null for Chinese, Japanese, Hindi, and Arabic. |
| key_field_accuracy | float | Proportion of structured ground-truth fields matched verbatim (0.0 to 1.0). |
| numeric_exact | float | Exact substring accuracy across monetary amounts, dates, times, and telephone numbers. |
| verdict | string | Evaluation outcome: PASS (≥95%), WARN (85–94.9%), or FAIL (<85%). |
Terms of Use & License
This benchmark dataset is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
You are free to share, copy, and adapt this data for research, commercial evaluation, and publication, provided you credit CreatorPrivacyKit Labs and cite the benchmark release.
Cite this benchmark
Empirical research published by CreatorPrivacyKit Labs is freely citeable with attribution under CC BY 4.0.
CreatorPrivacyKit Labs. (2026). Multilingual Browser OCR Benchmark Dataset (Version 1.0) [Data set]. CreatorPrivacyKit. https://creatorprivacykit.com/labs/ocr-benchmark/dataset/View BibTeX entry
@misc{creatorprivacykit_2026_ocr_benchmark,
title = {Multilingual Browser OCR Benchmark Dataset},
author = {{CreatorPrivacyKit Labs}},
year = {2026},
version = {1.0},
howpublished = {\url{https://creatorprivacykit.com/labs/ocr-benchmark/dataset/}},
note = {Tested 2026-09-20. In-browser Tesseract.js 5.1.1 multilingual evaluation}
}