---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: text
    dtype: string
  splits:
  - name: train
    num_bytes: 1759539306
    num_examples: 2160000
  download_size: 1866820804
  dataset_size: 1759539306
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
language:
- ar
---
# Arabic OCR Dataset

## Overview
The Arabic OCR Dataset is a comprehensive resource aimed at enhancing Optical Character Recognition (OCR) capabilities for the Arabic language. The dataset consists of over 2 million labeled images of Arabic text extracted from diverse sources, ideal for training and benchmarking Arabic OCR models.

## Dataset Details
- **Dataset Size**: ~2.16 million labeled samples
- **Total File Size**: 1.87 GB
- **Format**: Parquet
- **Modalities**: Images and Text
- **Languages**: Arabic

## Structure
Each entry in the dataset includes:
- `image`: An image file containing Arabic text.
- `text`: Corresponding Arabic text as ground truth.

The images vary in width from 29px to 222px, containing text samples ranging from 7 to 10 characters in length.

## Intended Uses
This dataset is designed for:
- Training state-of-the-art Arabic OCR models.
- Evaluating performance of OCR systems.
- Research in Arabic Natural Language Processing (NLP).

## Limitations
- Text length is limited to short to medium-length Arabic text snippets.
- Variability in image quality may affect OCR performance.

## How to Use
### Loading the Dataset
```python
from datasets import load_dataset

# Load Arabic OCR Dataset
dataset = load_dataset("mssqapi/Arabic-OCR-Dataset")
```

### Accessing Data Samples
```python
# Example of accessing data sample
sample = dataset['train'][0]
print(sample['text'])
display(sample['image'])
```