ios

MonOCR iOS

MonOCR iOS provides high-performance, native optical character recognition for the Mon script natively on Apple devices.

For mission context, community guidelines, and cross-platform information, please refer to the MonOCR Root Documentation.

Overview

By leveraging a Neural Engine optimized execution model, MonOCR iOS utilizes Core ML and the Vision framework to perform all character recognition on-device. This architecture is purpose-built for the Apple Neural Engine (ANE), ensuring high-throughput, low-latency performance while maintaining an absolute privacy model—guaranteeing that linguistic assets remain within the localized secure environment.

Key Features

• On-Device Inference: Optimized Core ML execution with ANE acceleration.
• Privacy by Design: Zero data collection; OCR processing is 100% local.
• Mon Language Support: Specialized for the Mon script (315-char charset).
• Line Segmentation: Automatic horizontal projection profiling for accurate text extraction.
• Modern UI: 100% SwiftUI with native animations and light/dark theme support.
• Format Support: Handles high-resolution images and multi-page PDFs.
• Script Fidelity: Integration of PyidaungSu fonts for correct Mon/Myanmar rendering.

Tip

File size is limited to 50MB for web and 20MB for mobile. For processing larger files or leveraging more powerful hardware, please use the CLI or package directly via uv add monocr or pip install monocr.

Architecture

Image (UIImage)
  LineSegmenter     -> horizontal projection profile -> List<LineSegment>
  ImagePreprocessor  -> grayscale + normalize [-1.0, 1.0]
  MonOcrEngine      -> Core ML Prediction (monocr.mlpackage)
  CtcDecoder        -> greedy CTC decode -> String

Model Specification

Attribute	Specification
Architecture	MobileNetV3 + BiLSTM-384 + CTC
Precision	FP32 (Core ML)
Parameters	~6.6M
Input	128 × Variable (H × W)
Asset Size	~23 MB

Project Structure

apps/ios/
├── monocr-ios/
│   ├── engine/           # OCR Core (Core ML, Preprocessing, Decoding)
│   ├── ui/               # SwiftUI Views & ViewModels
│   ├── persistence/      # SwiftData models & History
│   ├── resources/        # Models, Fonts, & Assets
│   └── util/             # Platform & PDF utilities
├── monocr-iosTests/      # Unit & Performance Tests
└── monocr-ios.xcodeproj  # Xcode Project

Ecosystem

MonOCR is a unified cross-platform ecosystem designed for parity and performance:

• MonOCR Web: Privacy-first in-browser OCR.
• MonOCR Android: (In this Monorepo) Native Jetpack Compose app.
• MonOCR iOS: (In this Monorepo) Native SwiftUI app.

Development

Prerequisites

• Xcode 15.0+
• iOS 16.0+
• Swift 5.9+

Getting Started

1. Clone the repository and open monocr-ios.xcodeproj in Xcode.
2. Ensure monocr.mlpackage is correctly referenced in the app target.
3. Build and run on a physical device for optimal performance (ANE acceleration).
4. Use the Document Picker or Camera to begin character recognition.

Resources

• Hugging Face Models (ONNX, TFLite, Core ML)
• Unified SDKs (ONNX Core)
• MonOCR Monorepo

Contributors

• Janakh Pon
• Oung Seik Nyan
• Rajel Da Key

License

MIT