chuuhtetnaing/myanmar-ocr-dataset-for-vlm

Myanmar OCR Dataset

A synthetic OCR dataset for fine-tuning Vision Language Models (VLMs) on Myanmar (Burmese) text recognition. It contains page images paired with their ground-truth text, sourced from chuuhtetnaing/mm-lib-book-dataset and rendered into page images using various Myanmar fonts.

Subsets

Subset	Description	Details
single_font	Rendered with Pyidaungsu font only	437 books
multi_font	Rendered with 76 Myanmar fonts	3 books (ပဋ္ဌာန်းမြတ်ဒေသနာ၊ ရတနာသုံးပါးကိုကိုးကွယ်ဆည်းကပ်လာသူတစ်ယောက်အကြောင်း၊ နဂါးနိုင်ဓါး) across all fonts

Each subset has train and test splits.

Usage

from datasets import load_dataset

# Load a specific subset and split
ds = load_dataset("your-username/myanmar-ocr-dataset", "single_font", split="train")

# Load all splits of a subset
ds = load_dataset("your-username/myanmar-ocr-dataset", "multi_font")
ds["train"]
ds["test"]

Each example has two fields:

• image: the page image
• text: the ground-truth text

Train/Test Split Strategy

5% of pages from each book are held out as the test set (at least 1 page per book).

Subset	Books	Split method
single_font	ပဋ္ဌာန်းမြတ်ဒေသနာ၊ ရတနာသုံးပါးကိုကိုးကွယ်ဆည်းကပ်လာသူတစ်ယောက်အကြောင်း၊ နဂါးနိုင်ဓါး	Last 5% sequentially
single_font	Remaining 434 books	Random 5% (seed=42)
multi_font	ပဋ္ဌာန်းမြတ်ဒေသနာ၊ ရတနာသုံးပါးကိုကိုးကွယ်ဆည်းကပ်လာသူတစ်ယောက်အကြောင်း၊ နဂါးနိုင်ဓါး	Last 5% sequentially

The three books above appear in both subsets. Their test pages are split sequentially in both cases to ensure the same pages are consistently held out regardless of the font used to render them.