README.md

metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: image
      dtype: image
  splits:
    - name: train
      num_bytes: 14791432
      num_examples: 64
  download_size: 14674791
  dataset_size: 14791432
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-nc-sa-4.0
language:
  - or
pretty_name: Odia OCR Dataset
size_categories:
  - n<1K

Dataset Card for Odia Lipi OCR Dataset

Dataset Description

This dataset is part of the Odia Lipi Project, an open and collaborative initiative led in collaboration with AHRC IIT Bhubaneswar, aimed at building high-quality OCR resources for the Odia language. It contains scanned page images paired with human-validated Odia text, intended to support OCR model training, post-OCR correction, NLP research, and the digital preservation of Odia literature and documents.

The dataset currently focuses on accurately digitizing printed Odia text and is designed to be extended to handwritten sources and multimodal OCR settings in future releases.

Uses

This dataset is intended for:

• Training and evaluating OCR models for the Odia language
• Post-OCR text correction and alignment tasks
• NLP research in low-resource Indian languages
• Digitization and preservation of Odia books, manuscripts, and archival documents

The dataset may also be used for academic research and non-commercial applications focused on language technology and cultural preservation.

Contributors

• Siba Prasad Sahu
• Manas Behera
• Satyam Swarup Rout
• Sk Shahid
• Sambit Sekhar
• Ganeshan K

Acknowledgement

We would like to thank the Odia Bibhaba team (https://odiabibhaba.in/) for curating old Odia literature, including books, magazines, and artifacts, which has been instrumental in building this open-source OCR dataset.

Citation

If you find this repository useful, please consider giving 👏 and citing:

@misc{odia_lipi_ocr_2026,
  author = {
    Siba Prasad Sahu and
    Manas Behera and
    Satyam Swarup Rout and
    Sk Shahid and
    Sambit Sekhar and
    Ganeshan K
  },
  title = {Odia Lipi: An Open-Source OCR Dataset for the Odia Language},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face Datasets},
  howpublished = {\url{https://huggingface.co/OdiaGenAI}},
}

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.