Skip to content
Dataset By Humans
On requestSoutheast Asia

Vietnamese Scene Text Image Dataset for OCR

Real shop signs, menus, banners and street notices in Vietnamese, full of diacritics, mixed fonts and imperfect lighting.

See pricing · from $490Get free samples

On request: this dataset is collected to your specification after you order. Typically 2–3 weeks for 1,000 images; transcriptions add time.

Pricing

Fixed prices for standard bundles, collected to your spec. Pick a package and a license to see the exact total, or ask for a custom quote.

Bundle packages compared
FeatureStarterProCustom
Images5002,000Any quantity
Metadata and topic labels✓Included✓Included✓Included
Human-written captions–Not included✓IncludedOptional
Boxes, masks, transcriptions–Not included–Not includedOptional
Datasheet, license, checksums✓Included✓Included✓Included
Research license$490$0.98 / image$1,490$0.75 / imageQuote
Commercial license$1,250$2.50 / image$3,900$1.95 / imageQuote
Exclusive license$4,900$9.80 / image$14,900$7.45 / imageQuote

How ordering works

  1. 1Choose a package and license, then pay a 50% deposit at checkout.
  2. 2Within 48 hours we confirm the spec and send 10–20 sample photos before full collection.
  3. 3We collect and deliver in parts so you can review early. Typically 2–3 weeks for 1,000 images; transcriptions add time.
  4. 4Pay the balance and download the final bundle.

Custom quote

Need another quantity, an Enterprise license or annotations for this topic? Available here:

  • Topic labels
  • Full-image transcriptions
  • Word-level boxes with transcription
Request a custom quote →

Overview

Vietnamese uses the Latin alphabet with stacked diacritics, so a single character can carry two marks. OCR models trained mostly on English text routinely drop or confuse these marks, especially on hand-painted signs, curved banners and low-light menus.

This dataset contains photos of Vietnamese text as it appears in daily life: storefront signs, restaurant menus, price boards, banners, notices and packaging, photographed at natural angles with real-world blur, glare and occlusion.

Sample photos

Real photos from this topic, shown at reduced resolution. Full-resolution samples are emailed on request.

  • Bilingual parking sign reading 'Toà/Building BS11 BS12' with an EV charging station sign, in front of high-rise apartments
  • Night street view of a print and photocopy shop with large Vietnamese signs reading 'Nhà in - Photocopy Nam Dũng' and parked motorbikes
  • Shopfront of a home-appliance company with dense Vietnamese signage, neighbouring food shop signs and motorbikes on the pavement
  • Corner snack store with large red and white Vietnamese signs, slogans and a hotline number, with street traffic below

Get free full-resolution samples

We email you a sample pack for this topic.

Used only to send your samples. Privacy policy.

Technical specifications

Capture device
Modern smartphones (device model recorded per image in EXIF)
File format
JPEG (original HEIC/RAW on request)
Resolution
Native sensor resolution, typically 12 MP or higher
Text types
Shop signs, menus, price boards, banners, notices, packaging
Language
Vietnamese, with some mixed Vietnamese–English text
Metadata
metadata.jsonl with device, exposure, timestamp, dimensions, topic labels
Privacy processing
Faces and license plates blurred, GPS removed
License
Research, Commercial, Enterprise or Exclusive

Use cases

  • Scene text recognition

    Fine-tune OCR and text-spotting models (PaddleOCR, TrOCR, docTR) to read Vietnamese diacritics accurately.

  • Multilingual VLM evaluation

    Benchmark how well vision-language models read and translate Vietnamese text in natural images.

  • Translation & accessibility apps

    Train camera-translation and screen-reader features for travellers and visually impaired users.

  • Retail & mapping

    Extract business names and categories from storefront imagery for maps and local search.

How this data is collected

  1. 1Captured on a smartphone by a person
  2. 2Duplicates and near-duplicates removed
  3. 3Blurry and broken images filtered out
  4. 4Faces and license plates detected and blurred
  5. 5Every image reviewed by a person
  6. 6GPS stripped; device EXIF kept
  7. 7Labels, captions and metadata.jsonl generated
  8. 8Datasheet and license packaged with the delivery
Read our full process →

Provenance & legal

  • 100% captured by people: no scraping, no generative AI
  • Datasheet documenting collection, processing and limitations
  • Commercial license that lets you keep models trained on the data
  • Supports training-data documentation under the EU AI Act
Licensing options →

Topic-specific notes

  • • Phone numbers and personal names on signs can be blurred on request; business names are kept.
  • • Bystanders' faces are blurred and children are excluded.
  • • Text transcriptions are written by native Vietnamese speakers.

Example metadata record

One line of metadata.jsonl per image. Fields vary with the annotation options you choose; values shown are illustrative.

{
  "id": "vst-000318",
  "file": "images/vst-000318.jpg",
  "topic": "scene-text",
  "text_type": "menu",
  "captured_at": "2026-09-20T19:44:02+07:00",
  "city": "Ho Chi Minh City, VN",
  "device": {
    "make": "Samsung",
    "model": "Galaxy S24"
  },
  "width": 4000,
  "height": 3000,
  "transcription": "PHỞ BÒ TÁI 45.000đ | PHỞ GÀ 40.000đ | BÚN CHẢ 50.000đ",
  "privacy": {
    "faces_blurred": 0,
    "phone_numbers_blurred": 1,
    "gps_removed": true
  }
}

Each delivery contains:

vietnamese-scene-text-ocr-v1/
├── images/
├── metadata.jsonl
├── DATASHEET.md
├── LICENSE.pdf
└── checksums.sha256

Frequently asked questions

Are transcriptions included?

Transcriptions are available on request, written by native Vietnamese speakers. Word-level bounding boxes with text can also be quoted.

Can you include handwritten text?

Yes, hand-painted signs and handwritten menus can be collected. Mention this in your request.

Is the text in the images real?

Yes. Every image is a photo of real signage or print. Nothing is rendered or generated.

How does payment work for a bundle?

You pay a 50% deposit at checkout and the balance when we deliver the final bundle. Free sample photos are available above before you order, and the deposit is refunded in full if we cannot deliver the agreed spec.

Why is a bundle so much cheaper per image than a single image?

A bundle is planned and collected as one job, so capture, review and privacy processing cost far less per photo. Single images are priced for buyers who need just a few specific photos.

What is the difference between non-exclusive and Exclusive?

With Research or Commercial we may license the photos collected for your order to other buyers later. With Exclusive they are never licensed to anyone else or added to our catalog.

Can I order a different quantity or extra annotations?

Yes. Request a custom quote with the quantity, annotations and deadline you need. Prices per image fall as quantity grows.

Need a variation?

Different angles, lighting, locations, quantity or labels? Describe it and we reply within 48 hours with matching samples and a quote.

Request a quote →

Related datasets