---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: text
    dtype: string
  splits:
  - name: train
    num_bytes: 4962857995.17
    num_examples: 29591
  download_size: 4963338712
  dataset_size: 4962857995.17
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: cc-by-nc-4.0
task_categories:
- image-to-text
language:
- km
tags:
- ocr
- khmer
- visual-grounding
pretty_name: Wild KhmerST
---


# Wild KhmerST Dataset (OCR)

This dataset is a high-quality collection of Khmer text images "in the wild".

## 📌 Provenance & Credits

- **Original Source:** [Khmer Word Dataset (Kaggle)](https://www.kaggle.com/datasets/keosaly/wildkhmerst-dataset)
- **Original Owner:** **Saly KEO**
- **Modification:** The original dataset (which provides labels per image) has been processed to create pairs of 'text region image' and 'text'. So, these pairs of 'text region image' and 'text' can be directly used to train or fine-tune OCR models.

## 📄 License

This dataset is distributed under the **Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)** license.

- **Attribution:** You must give appropriate credit to Saly KEO (original source) and the current maintainer.
- **Non-Commercial:** You may not use the material for commercial purposes.

## Dataset Summary

Each row is a pair of 'text region image' in `Image` object of `pillow` package and 'text' in `str`.

- **Primary Language:** Khmer (UTF-8)

## Dataset Structure

| Column  | Type     | Description                                                                          |
| :------ | :------- | :----------------------------------------------------------------------------------- |
| `image` | `Image`  | The image contains only text                                                         |
| `text`  | `String` | Text in the image                                                                    |

__NOTE:__ some images contain characters that are not included in their corresponding 'text' column (_i.e.,_ this dataset contains some noise). This is because original dataset provides polygon text region and rectangular region that superscribe the polygon is used to crop the image.

## Citation

If you use this dataset, please credit the original author:

```text
Original Creator: Saly KEO (Kaggle)
Derivative Work: Kao Vichet
License: CC BY-NC 4.0
```