Datasets:

acul3
/

WikiCitra-ID

dataset_info:
  features:
    - name: url_ori
      dtype: string
    - name: captions
      sequence: string
    - name: metadata
      list:
        - name: wikipedia_url
          dtype: string
        - name: caption
          dtype: string
        - name: url_image
          dtype: string
    - name: caption_count
      dtype: int64
  splits:
    - name: train
      num_bytes: 148962879
      num_examples: 352150
  download_size: 60409774
  dataset_size: 148962879
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

WikiCitra-ID Dataset

WikiCitra-ID is a large-scale image-text dataset collected from Indonesian Wikipedia. The dataset consists of images and their corresponding captions extracted from the latest HTML dump of Indonesian Wikipedia articles.

Dataset Description

WikiCitra-ID contains paired image-text data sourced from Indonesian Wikipedia articles. This dataset is derived from the image portion of WikipediaMM-Indo and has been processed to create clean image-text pairs.

Statistics

Total number of images: 352,150
Number of unique images with non-empty captions: 325,971

Data Format

Each entry in the dataset contains:

Image: Visual content from Wikipedia articles
Text: Corresponding Indonesian caption or description
Metadata: Additional information for verification

Limitations and Considerations

Please be aware of the following limitations when using this dataset:

Icon and Symbol Presence
- The dataset may contain non-content images such as:
  - Directional arrows
  - National symbols
  - UI elements
  - Other Wikipedia icons
Caption Quality
- Some image-caption pairs may show inconsistencies due to crawling limitations
- Users can verify the alignment using provided metadata
- We recommend implementing additional filtering based on your specific use case

Usage

This dataset is suitable for:

Multimodal machine learning research
Indonesian language vision-language models
Cross-lingual image understanding
Wikipedia-based knowledge extraction

Citation

If you use this dataset in your research, please cite:

@misc{wikicitra2024,
  title={WikiCitra-ID: A Large-scale Indonesian Wikipedia Image-Text Dataset},
  year={2024},
  publisher={Hugging Face}
}

License

This dataset follows the same licensing as Wikipedia content. Images and text are subject to their original licenses on Wikipedia.

Acknowledgments

Original data sourced from Indonesian Wikipedia
Built upon the WikipediaMM-Indo dataset

Contact

For questions and feedback about the dataset, please open an issue in this repository.