---
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
---

# Arabic-Image2Html Dataset

A dataset of **28K image-HTML pairs** for training OCR models to transform Arabic documents into structured HTML.

## Dataset Description

This dataset contains Arabic document images paired with semantic HTML output. It includes diverse document types and layouts for training image-to-HTML models.

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("abdouuu/arabic-image2html")
```

## Dataset Stats

- **Total samples:** 28,044
- **Train:** 22,435 examples
- **Validation:** 2,804 examples
- **Test:** 2,805 examples
- **Language:** Arabic
- **Output format:** Semantic HTML

## License

Apache 2.0