imagesWantedimageshandwritingocrhuman-annotated
Everyday handwriting: notes, lists and corrections
What they need
Shopping lists, reminders, quick instructions, crossed-out words. I'd like a small collection of the kind of handwriting people actually use day to day. The request is for 30 notes in English from three consenting adults, ten notes each, photographed or scanned and paired with exact transcriptions. Each note should be 10-30 words; at least six should include a correction or crossed-out word. Writing fresh notes for this is completely fine. Keep the spelling and punctuation as written, and ask each writer to confirm what they meant where something is hard to read. The detailed file and annotation requirements are below.
Requirements
formatzip
coverageExactly 3 adult writers and 10 different notes per writer. At least 6 notes overall must contain a crossed-out word or an inserted correction, with at least one such note from each writer. Use genuine handwriting, not fonts or digitally rendered text. Notes may be deliberately written for this task.
acceptance30 distinct, legible images with the required writer balance, faithful annotations, and no duplicated images or reused note content. Include all originals; cropping and rotation are permitted but alterations to handwriting are not.
deliverablesOne ZIP with 30 original JPG or PNG images, annotations.jsonl (one record per image), README.txt, and a participant-consent attestation using anonymous writer IDs. Each note must contain 10-30 words. Images must be in focus, show the entire note, and be at least 1200 pixels on the longest side.
required fieldsimage_id, file_name, writer_id, verbatim_transcription, final_intended_text, corrections (crossed-out text and replacement or insertion), uncertain_spans, capture_date
transcription rulesPreserve spelling, punctuation, capitalization, and line breaks. Record legible crossed-out text in the corrections field; mark genuinely unreadable text as [illegible] rather than guessing. The writer must confirm the final intended text, and the submitter must check each transcription against the image.
consent and provenanceWriters must consent to dataset redistribution and OCR research/training use. Use anonymous IDs and retain the underlying consent records; do not upload signatures, addresses, phone numbers, or private messages. Include a dated consent attestation, collection method, and reuse terms. No generated handwriting or existing handwriting dataset extracts.
$45.00
bounty reward per winner
0 submissions
Up to 1 winner
Posted by Shaurya Agarwal
Review target: 48 hours after a safe sample is ready.
20d 1h remaining