README.md

metadata

dataset_info:
  features:
    - name: image
      dtype: image
    - name: text
      dtype: string
    - name: style
      dtype: string
  splits:
    - name: train
      num_bytes: 532905094.304
      num_examples: 41664
  download_size: 266331659
  dataset_size: 532905094.304
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
language:
  - my
size_categories:
  - 10K<n<100K
license: cc-by-4.0
task_categories:
  - feature-extraction
tags:
  - ocr
  - image-to-text
pretty_name: MyanmarOCR-ImageText

🇲🇲 MyanmarOCR-ImageText Dataset

A clean and diverse Burmese Image-to-Text dataset for OCR and multimodal AI research.

---

📌 Summary

• Total images: 41,664
• Unique Burmese text entries: 1,139
• Styles per text: 32 variations each
• Resolution: 512 × 512
• File types: PNG/JPG images
• Dataset split: train only
• Use cases: OCR, I2T (image-to-text), VLM pretrain/fine-tune

All text is Burmese only.
No English words and no punctuation like: ? , ' " -

---

🔡 Text Content

Includes:

• Common words & Pali words
• Signs and short phrases
• Full Myanmar Unicode support

Variety of Myanmar spellings and writings are included
(မြန်မာအက္ခရာတွေနဲ့ စကားလုံးမျိုးစုံပါဝင်ပါတယ်)

---

🎨 Style Variations

Each text appears in 32 visual styles with differences in:

• font
• color
• texture
• rotation (small angle)
• background patterns

This helps models generalize across real-world environments.

---

🧩 Data Format

Each sample includes:

Column	Type	Description
image	image	512×512 Burmese rendered text
text	string	Ground truth Burmese label
style	string	Style ID (e.g., style_01)

Example record:

{
  "image": "<image>",
  "text": "မြန်မာနိုင်ငံ",
  "style": "style_07"
}

---

🧪 Usage

from datasets import load_dataset

ds = load_dataset("kalixlouiis/MyanmarOCR-ImageText", split="train")
print(ds[0])

ds[0]["image"].show()

---

🎯 Intended Purposes

• Burmese OCR training
• Scene text model finetuning
• Vision-language pretraining
• Synthetic-to-real text recognition research
• Benchmark for Myanmar multimodal AI

---

⚠️ Limitations

• Synthetic images only — not real photos/signboards
• No English text or punctuation
• No complex layout structures (single word/short text per image)

---

📄 License

This dataset is released under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license.

You are free to:

• ✔ Share — copy and redistribute for any purpose
• ✔ Adapt — modify, transform, and build upon the data

As long as you:

• Give appropriate credit
• Indicate changes
• Provide a link to the license

📌 License Text: https://creativecommons.org/licenses/by/4.0/

---

✨ Acknowledgment

Created by @kalixlouiis
with the goal of improving Myanmar OCR and AI research.

If you use this dataset in your research or applications, please cite and provide a link to the dataset page on Hugging Face:

🔗 https://huggingface.co/datasets/kalixlouiis/MyanmarOCR-ImageText

---

📚 Citation

@dataset{kalixlouiis2025myanmarocr,
  title        = {MyanmarOCR-ImageText},
  author       = {Kalix Louis},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/kalixlouiis/MyanmarOCR-ImageText}},
  license      = {CC-BY-4.0}
}

---