Myanmar OCR Data Generator

A text recognition data generator with Myanmar language support, forked from TextRecognitionDataGenerator (trdg).

Overview

This repository extends the original TextRecognitionDataGenerator with specific support for Myanmar language. Key improvements include:

• Added Myanmar language support for synthetic text generation
• Fixed issues with text splitting specific to Myanmar language

Why This Fork

I've chosen to create a separate repository rather than a PR to the original project as the original repository hasn't been actively maintained recently.

Features

• All original TRDG features (multiple fonts, background, distortion, etc.)
• Myanmar language support with proper Unicode handling
• Correct text splitting for Myanmar characters
• Support for Myanmar dictionary-based word generation

Installation

# Clone the repository
git clone https://github.com/yourusername/myanmar-ocr-data-generator.git
cd myanmar-ocr-data-generator

# Install dependencies
pip install -r requirements.txt

Usage

Prerequisites

Before using the tool with Myanmar language, you need to generate the Myanmar dictionary file:

1. Run the sample dictionary generation notebook (notebook/generate-dictionary.ipynb) to generate the my.txt dictionary file
2. Place the my.txt dictionary file in trdg/dicts/

Command Line Interface

Generate random Myanmar text images:

cd trdg
python3 run.py -c 10 -l my

Generate specific Myanmar text from your text file:

python3 run.py -l my --dict my_word_list.txt -c 10

Python API

from trdg.generators import GeneratorFromStrings

# Generate images from a Myanmar dictionary
generator = GeneratorFromStrings(
    strings = ["ဂရိဒဏ္ဍာရီ", "သည်", "ရှေးခေတ်ဂရိလူမျိုးများ", "မူလအနေဖြင့်", "ပြောဆိုခဲ့ကြသော", "ဒဏ္ဍာရီ"],
    language='my',
    count=10
)

# Generate your text images
for img, lbl in generator:
    # Do something with the generated image and label
    img.save(f"output/{lbl}.jpg")

Basic (CLI)

python3 run.py -c 10 -w 5 -f 64 -l my

You get 1,000 randomly generated images with random text on them like:

By default, they will be generated to out/ in the current working directory.

Text skewing

What if you want random skewing? Add -k and -rk (python3 run.py -c 10 -w 5 -f 64 -k 5 -rk -l my)

Text distortion

You can also add distortion to the generated text with -d and -do (python3 run.py -c 10 -w 5 -f 64 -k 5 -d 1 -l my)

Text blurring

But scanned document usually aren't that clear are they? Add -bl and -rbl (python3 run.py -c 1 -w 5 -f 64 -bl 4 -l my) to get gaussian blur on the generated image with user-defined radius (here 0, 1, 2, 4):

Background

Maybe you want another background? Add -b (python3 run.py -c 1 -w 5 -f 64 -b 3 -l my) to define one of the three available backgrounds: gaussian noise (0), plain white (1), quasicrystal (2) or image (3).

When using image background (3). A image from the images/ folder will be randomly selected and the text will be written on it.

Generate Large Synthetics Data for PaddleOCR Training

Please use the following scripts and notebook under notebook/paddleocr to generate data for the PaddleOCR format.

• Run the generate-dictionary.ipynb notebook to generate corpus for training from chuuhtetnaing/wikipedia and facebook/flores datasets.
• 1. generate.py script to generate the synthetics OCR dataset.
• 2. create-lable-for-hf.py and 3. upload-to-huggingface.py scripts to create the lable for huggingface format and upload to huggingface. (If you don't wanna upload to huggingface, please skip these scripts)

Acknowledgments

• TextRecognitionDataGenerator for the original implementation