MonOCR (Universal SDK)

MonOCR is a high-performance, production-ready Optical Character Recognition (OCR) engine specifically optimized for the Mon language (mnw). Built on ONNX Runtime, it provides a lightning-fast, unified API for text recognition across any platform.

Why MonOCR?

• Production Accuracy: Aligned with the latest high-precision models (128px vertical resolution).
• Universal SDK: Native, high-performance implementations for JavaScript, Python, Go, and Rust.
• Robust Segmentation: Intelligent line-detection with adaptive thresholding and relative padding for varied document layouts.
• Smart Model Management: Zero-config setup; models are automatically fetched and cached from Hugging Face.

Supported Platforms

SDK	Directory	Registry/Source	Status
JavaScript / Node.js	js/	npm: monocr	Production
Python	python/	PyPI: monocr-onnx	Production
Go	go/	github.com/MonDevHub/monocr-onnx/go	Production
Rust	rust/	monocr-onnx	Production

Quick Installation

Python

pip install monocr-onnx

Node.js

npm install monocr

Go

go get github.com/MonDevHub/monocr-onnx/go

Usage Example (Python)

from monocr_onnx import MonOCR

# Initialize engine (downloads model automatically on first run)
engine = MonOCR()

# Simple page-level OCR
text = engine.predict("scanned_document.jpg")
print(f"Recognized Text:\n{text}")

# Or process specific lines if you have your own layout analysis
line_text = engine.predict_line("single_line_crop.png")

Project Structure

• python/: Source code for the Python package.
• js/: Source code for the Node.js package (uses sharp for image processing).
• go/: Source code for the Go module.
• rust/: Source code for the Rust crate.
• models/: (Reference) Model architecture and conversion scripts.

Model Hub

The underlying weights and multi-format exports (ONNX, MLPackage, PyTorch) are hosted on the Hugging Face Model Hub.

License

MIT License. Developed and maintained by the MonDevHub team.