Skip to content

README.md

Latest commit

 

History

History
123 lines (98 loc) · 3.03 KB

README.md

File metadata and controls

123 lines (98 loc) · 3.03 KB

HalalBench Dataset

Overview

The HalalBench dataset contains 1,043 food packaging images with 36,438 text annotations across 14 languages, annotated in COCO format with bounding boxes and transcriptions for ingredient text regions.

Download

The dataset will be available for download at:

https://github.com/halallens-no/halalbench/releases

After downloading, extract into this directory:

cd data/
tar -xzf halalbench-v1.0.tar.gz

This will produce:

data/
  annotations.json     # COCO-format ground truth
  images/              # 1,043 food packaging images
    img_0001.jpg
    img_0002.jpg
    ...

COCO Annotation Format

The annotations follow the standard COCO format with additional fields for OCR:

{
  "images": [
    {
      "id": 1,
      "file_name": "img_0001.jpg",
      "width": 1920,
      "height": 1080,
      "language": "en"
    }
  ],
  "annotations": [
    {
      "id": 1,
      "image_id": 1,
      "bbox": [x, y, width, height],
      "category_id": 1,
      "attributes": {
        "text": "sodium benzoate",
        "confidence": 1.0
      }
    }
  ],
  "categories": [
    {
      "id": 1,
      "name": "ingredient_text"
    }
  ]
}

Key Fields

Field Description
images[].language ISO 639-1 language code of the primary text
annotations[].bbox Bounding box in COCO format [x, y, w, h]
annotations[].attributes.text Ground-truth transcription of the text region

Language Distribution

Language Code Images Annotations
Arabic ar 72 2,518
Danish da 68 2,380
Dutch nl 71 2,485
English en 112 3,920
French fr 89 3,115
German de 85 2,975
Indonesian id 62 2,170
Japanese ja 78 2,730
Korean ko 74 2,590
Malay ms 59 2,065
Norwegian no 65 2,275
Swedish sv 70 2,450
Thai th 69 2,415
Turkish tr 69 2,350
Total 1,043 36,438

Annotation Guidelines

Each image was annotated by at least two annotators with the following protocol:

  1. Bounding boxes tightly enclose each ingredient text region
  2. Transcriptions preserve the original text as printed (including accents, special characters)
  3. Language labels reflect the primary language of the ingredient list on each image
  4. Inter-annotator disagreements resolved by a third annotator

License

The HalalBench dataset is licensed under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0).

You are free to share and adapt the dataset, provided you give appropriate credit and distribute any derivative works under the same license.

Full terms: https://creativecommons.org/licenses/by-sa/4.0/

Citation

@article{halalbench2026,
  title     = {HalalBench: A Multilingual OCR Benchmark for Food Packaging Ingredient Extraction},
  author    = {Nurul Isma},
  journal   = {arXiv preprint arXiv:XXXX.XXXXX},
  year      = {2026}
}