---
license: cc-by-nc-sa-3.0
dataset_info:
  features:
  - name: cocoid
    dtype: string
  - name: image
    dtype: image
  - name: caption
    dtype: string
  splits:
  - name: train
    num_bytes: 1606551321
    num_examples: 10000
  download_size: 1602120962
  dataset_size: 1606551321
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---
# Thai Image Captioning Dataset (Samples 10k)

This dataset is a **Thai image captioning corpus** created to generate natural and human-like image captions in Thai. 
It includes a curated subset of **10,000 samples** randomly selected from the original training data, providing a smaller yet representative dataset for efficient experimentation.

---

## Dataset Description

The dataset comprises high-quality image captions in Thai, designed for training and evaluating image captioning models. It integrates image-caption pairs from two main sources:  
1. **Microsoft COCO** – A publicly available dataset enriched with English captions.  
2. **Nectec IPU24** – A proprietary collection of image-caption pairs curated by NECTEC (Thailand).

The smaller version, provided here (`train-sample10k`), is derived from the original datasets, offering a compact and effective resource for research and development.

---

## Data Source and Download

- The full COCO 2017 dataset can be accessed and downloaded from the official website: [COCO Dataset](https://cocodataset.org/#download).  
- The Nectec IPU24 dataset is available through NECTEC's official portal: [NECTEC Corpus](https://aiforthai.in.th/corpus.php).

---

## Licensing and Usage

The dataset inherits licensing terms from its respective sources:
1. **COCO dataset** is licensed under the [Creative Commons Attribution 4.0 License](https://creativecommons.org/licenses/by/4.0/).  
2. **NECTEC Dataset** is licensed under the [Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License](https://creativecommons.org/licenses/by-nc-sa/3.0/).

Please ensure proper attribution when using this dataset in research or commercial applications.