SoyVitou/KhmerST-Dataset-OCR-Cropped

KhmerST OCR Cropped Dataset

This repository provides cropped scene-text images for OCR training, derived from the KhmerST benchmark dataset.

The dataset is intended for Khmer Scene Text Detection and Recognition research, especially for low-resource OCR tasks.

---

✅ Credits / Citation

If you use this dataset, please cite the original KhmerST paper:

KhmerST: A Low-Resource Khmer Scene Text Detection and Recognition Benchmark
Vannkinh Nom, Souhail Bakkali, Muhammad Muzzamil Luqman, Mickaël Coustaty, Jean-Marc Ogier

📄 Paper: https://arxiv.org/pdf/2410.18277

Original dataset source:
https://gitlab.com/vannkinhnom123/khmerst

---

Dataset Description

KhmerST is the first Khmer scene-text dataset, containing:

• 1,544 expert-annotated images
  • 997 indoor scenes
  • 547 outdoor scenes
• Diverse conditions:
  • flat and raised text
  • low illumination
  • distant and partially occluded text
• Line-level text annotations
• Polygon bounding boxes

This cropped version provides text-line images ready for OCR model training.

---

Dataset Format

Each sample contains the following columns:

Column	Type	Description
id	int	Unique identifier
image	Image	Cropped text-line image (previewable in HF viewer)
label	string	Khmer transcription text
filename	string	Relative image path

Example:

{
  "id": 0,
  "image": "img/sample_01.jpg",
  "label": "សូមអរគុណ",
  "filename": "img/sample_01.jpg"
}

Usage

You can load the dataset with:

from datasets import load_dataset

ds = load_dataset("SoyVitou/KhmerST-Dataset-OCR-Cropped")
print(ds["train"][0])

Acknowledgment

All credit goes to the KhmerST dataset creators

Vannkinh Nom, Souhail Bakkali, Muhammad Muzzamil Luqman,
Mickaël Coustaty, and Jean-Marc Ogier.