Please visit to the GitHub repository for other Myanmar Langauge datasets.

myOCR-hf

This is a preprocessed version of the myOCR dataset in HuggingFace image dataset format.

Changes from Original

• Converted from tab-separated format to structured dataset with image and text columns
• Properly formatted as HuggingFace Image dataset for easy loading and use
• Converted word-segmentation delimiters from underscores (_) to spaces

Dataset Structure

DatasetDict({
    train: Dataset({
        features: ['image', 'text'],
        num_rows: 18053
    })
    validation: Dataset({
        features: ['image', 'text'],
        num_rows: 5803
    })
})

Example:

• Original: Images/Press_1127.png\tအားနာ_စရာ_ကြီး_အကြာကြီး_ရှာ_နေ_ရ_တယ်
• Preprocessed:
  • image: PIL Image from Images/Press_1127.png
  • text: အားနာ စရာ ကြီး အကြာကြီး ရှာ နေ ရ တယ်

Usage

from datasets import load_dataset

dataset = load_dataset("chuuhtetnaing/myOCR-hf")

# Access image and text
example = dataset['train'][0]
image = example['image']  # PIL Image
text = example['text']    # Burmese text with spaces

Preprocessing

See the data preparation notebook for details.

Citation

Please cite the original dataset:

Thura Aung, Ye Kyaw Thu and Myat Noe Oo, "myOCR: Optical Character Recognition for Myanmar language with Post-OCR Error Correction", In Proceedings of the 19th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP 2024), Nov 11 to 15, 2024, Pattaya, Thailand.

@INPROCEEDINGS{10799448,
  author={Aung, Thura and Thu, Ye Kyaw and Oo, Myat Noe},
  booktitle={2024 19th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP)}, 
  title={myOCR: Optical Character Recognition for Myanmar language with Post-OCR Error Correction}, 
  year={2024},
  volume={},
  number={},
  pages={1-6},
  keywords={Translation;Optical character recognition;Bidirectional long short term memory;Transformers;Optical imaging;Feature extraction;Adaptive optics;Natural language processing;Convolutional neural networks;Iterative decoding;Optical Character Recognition;Myan-mar language;Synthetic data;Post-OCR;low-resource},
  doi={10.1109/iSAI-NLP64410.2024.10799448}}
}

Original Dataset

• Original Dataset: LULab/myOCR
• Github Repo: ye-kyaw-thu/myOCR
• Paper: "myOCR: Optical Character Recognition for Myanmar language with Post-OCR Error Correction" (iSAI-NLP 2024)

License

Same as the original dataset.