README.md

---
license: mit
task_categories:
- image-to-text
- object-detection
- token-classification
language:
- id
- en
tags:
- receipt
- ocr
- information-extraction
- cord
- indonesian
size_categories:
- n<1K
dataset_info:
  features:
  - name: image
    dtype: image
  - name: ground_truth
    dtype: string
  splits:
  - name: train
    num_bytes: 7311152.0
    num_examples: 5
  download_size: 7282064
  dataset_size: 7311152.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# parlarlax/tiny-cord


CORD (Consolidated Receipt Dataset) is a dataset for receipt understanding tasks.
This dataset contains Indonesian restaurant receipts with structured annotations
for menu items, prices, and text extraction with bounding boxes.


## Dataset Details

### Dataset Description

The CORD dataset contains receipt images and their corresponding structured annotations.
Each example includes:

- **Receipt Image**: High-resolution image of Indonesian restaurant receipts
- **Menu Items**: Structured data with item names, quantities, and prices  
- **Totals**: Subtotal, service charges, taxes, and final total
- **Text Annotations**: Detailed text extraction with bounding box coordinates

### Dataset Structure

```python
{
    'image': PIL.Image,
    'image_id': int,
    'image_size': {'width': int, 'height': int},
    'version': str,
    'split': str,
    'menu_items': [
        {'nm': str, 'cnt': str, 'price': str}, ...
    ],
    'totals': {
        'subtotal_price': str,
        'service_price': str, 
        'tax_price': str,
        'etc': str,
        'total_price': str
    },
    'text_annotations': [
        {
            'words': [{'text': str, 'bbox': [int, int, int, int], 'is_key': int}, ...],
            'category': str,
            'group_id': int,
            'sub_group_id': int
        }, ...
    ]
}
```

### Supported Tasks

- **Receipt Understanding**: Extract structured information from receipt images
- **OCR (Optical Character Recognition)**: Text extraction with spatial information
- **Information Extraction**: Named entity recognition for receipt components
- **Document Layout Analysis**: Understanding spatial relationships in receipts

### Languages

The receipts contain text in:
- Indonesian (primary language)
- English (some menu items and labels)

### Dataset Statistics

- Number of examples: Varies based on available receipt images
- Image dimensions: 864 x 1296 pixels
- Average menu items per receipt: ~20-25 items
- Text annotations include bounding boxes for precise localization

## Dataset Creation

This dataset was created from receipt images and corresponding JSON annotations
containing ground truth information about menu items, prices, and text locations.

### Source Data

The source receipts are from Indonesian restaurants, primarily from the Bali region.
All prices are in Indonesian Rupiah (IDR).

## Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("parlarlax/tiny-cord")

# Access an example
example = dataset['train'][0]
image = example['image']
menu_items = example['menu_items']
total_price = example['totals']['total_price']
```

## Dataset Card Contact

For questions or issues regarding this dataset, please create an issue in the repository.