Multilingual handwritten form OCR data
Forms filled in by hand the way people really fill them in: crowded boxes, crossed-out letters, numbers that spill over the line, two scripts on one page. Written by native writers in Latin, Devanagari, Arabic and East Asian scripts, and transcribed field by field, with every handwritten value matched to its printed label.
Specification
- Scripts
- Latin (English, Spanish and more), Devanagari (Hindi, Marathi), Arabic, Chinese, Japanese and Korean
- Documents
- Application, registration, intake, delivery and survey forms, and handwritten notes on printed pages
- Images
- Phone photos and phone scans, JPEG, PNG or PDF
- Annotation
- Field boxes, printed key and handwritten value pairs, full transcription, checkboxes and crossed-out text
- Privacy
- Personal details made up by the writers or replaced before delivery
- Quality check
- Every form is read by several native reviewers of its script
Example of the data
What you see, and the file you get.
- Field
- नाम (Name)
- Value
- प्रिया शर्मा
- Script
- Devanagari · Hindi
- Checkbox
- ticked: Married
| image_url | script | key | value | crossed_out | box |
|---|---|---|---|---|---|
| …/images/form_0001.jpg | devanagari | नाम | प्रिया शर्मा | 220 310 640 70 | |
| …/images/form_0001.jpg | latin | Date of birth | 12/04/1987 | 1978 | 220 402 380 70 |
| …/images/form_0002.jpg | japanese | 氏名 | 山田 花子 | 180 260 560 80 | |
| …/images/form_0003.jpg | arabic | الاسم | ليلى حسن | 760 300 520 72 | |
| …/images/form_0004.jpg | korean | 성명 | 김민지 | 200 280 480 76 |
Sample rows. Also as JSON or Excel.
Contact us to get the dataset
Tell us what you need. We walk you through the data, its formats and how it is checked on a short call.
What it is used for
Handwriting recognition
Train OCR that reads handwriting in many scripts, not only English.
Form processing
Extract handwritten values from application and intake forms.
Document AI evaluation
Test vision-language models on messy, real handwriting.
Records digitisation
Turn handwritten paper archives into searchable data.
Questions
Which scripts and languages are covered?
Latin, Devanagari, Arabic, Chinese, Japanese and Korean, written by native writers; tell us the languages and form types you need.
Are crossed-out and corrected entries kept?
Yes. Crossed-out text, corrections and ticked boxes are labelled, so models learn which value counts.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples