Street and shop signs with text for OCR training
OCR Data · India

OCR Training Data
Collection in India

Collect photo–transcription pairs from India: receipts, shop signs, menus and labels in Indic scripts and Indian English. You write the capture rules and approve every pair.

Ground truth in Indic scripts

Photo plus exact typed text — Devanagari, Tamil, Bengali, ₹ and mixed lines.

Wild capture, not clean scans

Thermal receipts, kirana tags and shop boards shot on phones.

You gate every pair

Compare image to transcription. Pay only when the pair is exact.

200+

Countries

13,487

Tasks posted

195,000+

Workers available

~4 Hours

Average time to first results

Why India

What Breaks OCR on Indian Text in the Wild

Latin-only, studio-scan training sets fail on Indic shop text and thermal receipts. Collect the pairs that match production photos.

Two scripts on one board

Hindi + English, Tamil + English, Bengali + Latin — the line your model must read is often mixed, not a textbook sample.

Thermal and handwritten prices

Faded kirana receipts and chalk or pen tags break models trained on clean PDFs and Western invoices.

Phone glare and harsh sun

Glass boards, noon light, shade under awnings. Collect the capture conditions users actually submit.

Native typists as labels

Contributors type what they see — conjuncts, ₹, abbreviations — so ground truth matches the photo.

Scripts

Name the Script in the Brief

India is the country filter. The script you need belongs in the instructions — then reject pairs that use the wrong one.

  • हिन्दी

    Hindi

  • বাংলা

    Bengali

  • తెలుగు

    Telugu

  • मराठी

    Marathi

  • தமிழ்

    Tamil

  • اردو

    Urdu

  • ગુજરાતી

    Gujarati

  • ಕನ್ನಡ

    Kannada

  • മലയാളം

    Malayalam

  • ଓଡ଼ିଆ

    Odia

  • ਪੰਜਾਬੀ

    Punjabi

  • অসমীয়া

    Assamese

  • भोजपुरी

    Bhojpuri

  • English

    Indian English

  • Hinglish

    Hindi + English

Start with a 20-pair pilot for less common scripts. Extend the same task once the transcriptions look right.

Sources

Where the Text Lives in India

Pick the surface your model fails on. Contributors shoot that surface on a phone — or fill your blank templates with fictional data.

  • Shop receipts

    Thermal kirana and café slips — faded, folded, half cut off.

  • Invoices and bills

    Local layouts, GST-style lines and mixed-language headers.

  • Menus

    Printed or laminated cards; Indic + English on the same page.

  • Price tags

    Handwritten ₹ amounts and shelf stickers under plastic.

  • Product labels

    Packaging text, ingredients and brand names on real SKUs.

  • Shop and street signs

    Painted boards, neon and printed notices in mixed scripts.

  • Handwritten notes

    Lists and ledgers — useful when print OCR is not enough.

  • Printed forms

    Filled checklists and applications; cover personal fields.

  • Template forms

    Your blanks, filled with fictional data — safest for PII.

Privacy

Documents Leak Data — Design the Brief for That

OCR collection is high-risk for personal data. The safe default is fictional templates; real receipts need cover-up rules and hard rejects.

Purpose before accept

State that people photograph text for OCR training. Contributors see it before they start.

No IDs. No bank docs.

Ban Aadhaar, PAN, passports and statements in the brief. Prefer your blank templates with fictional fields.

Cover or blur PII on real items

Names, phones and card numbers get covered before the shot. Reject leaks in review. You own DPDP compliance for anything that can identify a person.

One Brief

Same Brief, Different Scripts

One English brief. Contributors in different cities return the same task type — with local script on the photo and in the transcription.

Your brief

Photograph the nearest shop sign or price board. Type every character exactly — including ₹ and mixed English lines. No guessing faded text.

HindiDelhi

ताज़ा सब्ज़ी · ₹40

HinglishMumbai

Fresh Juice · ₹40

TamilChennai

புதிய காய்கறிகள்

BengaliKolkata

তাজা সবজি

  1. 01

    Write one brief: surface to shoot, script to type, exact-character rules.

  2. 02

    Set country to India; name city or script if you need a narrower mix.

  3. 03

    Reject pairs where the typed line does not match the photo.

How It Works

A Pair Is Not Optional

OCR training needs image and ground-truth text locked together. That is the unit you pay for and the unit you review.

  1. 01

    Define the pair

    Image collection, country India, surface to shoot, script to type. Require line-by-line typing that keeps ₹, digits and symbols exact.

  2. 02

    Capture and type together

    One submission = photo + transcription. Native readers type Indic text; you do not get image-only dumps.

  3. 03

    Gate on character match

    Approve only when typed text matches the photo. For stricter QC, run a second task that checks pairs.

Printed menu or document photographed for OCR training
Typed transcription
THALI HOUSE
Veg Thali180
Masala Chai40
TOTAL220

Country: IndiaLanguage: Hindi, EnglishPhone camera

Compare

Bought Packs vs Paired Collection

OCR buyers choose between opaque labeled dumps and controlled photo–text pairs. RapidWorkers is the second path: you own the script rules and the review gate.

What to check

Labeled document pack

RapidWorkers

Unit of work

Pages or images with vendor labels

Photo + typed transcription you review

Indic scripts

Often a short fixed language list

Name Devanagari, Tamil, Bengali… in the brief

Start

Quote and minimum volume

Pilot from $0.30 the same day

Control

Vendor delivers a finished set

You reject bad pairs before payout

Privacy

Unclear how PII was handled

Your brief bans IDs; you see every file

Iterate

New SOW for more scripts

Extend the same task after the pilot

How We Work

Self-Serve Pairs or Managed Script Coverage

Keep character-level review yourself, or have our team run script mix, recruitment and QC in India.

Self-serve

You own the brief and every pair

Launch an India image task, require photo + exact typing, then accept or reject each pair yourself.

Create an OCR task for India

Fully managed

We cover scripts and check pairs

We plan Devanagari / Tamil / Bengali (or your list), recruit, QC transcriptions and deliver in your format.

Talk to us
Pricing

Pay for Pairs, Not Guesswork

A sign photo with three typed lines costs less than a multi-line receipt. Rate follows typing effort. The 15% platform fee is in the estimate.

?
?
min
?
$
FAQ

Frequently Asked Questions

A phone photo of the text plus a typed transcription in the same submission. You review the pair character by character before payout.

Devanagari, Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Odia, Gurmukhi, Urdu, plus Latin for Indian English. Name the script in the brief; coverage depends on who takes the task.

Yes. Ask contributors to type mixed lines exactly as printed. That is a common failure mode for OCR in India.

Ban IDs and bank documents. Prefer blank templates filled with fictional data. On real receipts, require covered names and card numbers — and reject leaks.

Country is India. Put city, state or script in the instructions and reject mismatches. Pilot first for less common scripts.

Reject imperfect pairs yourself, then optionally launch a second task where other workers compare photo and text.

You set pay from $0.10 plus 15% fee. Full receipt typing costs more than a short sign. Campaigns start from $0.30.

Yes. We plan the script mix, recruit, QC pairs and deliver in your format.