English | မြန်မာဘာသာ | ဘာသာမန်
Mon is spoken by roughly one million people across Myanmar and Thailand. UNESCO classifies it as vulnerable — and no OCR toolchain existed for it before this project.
MonOCR takes an image of Mon script and returns text. It runs on Web, Android, and iOS — fully offline, no data leaves the device.
Built and maintained by the Mon developer community.
- Web: ocr.mondevhub.com
- Android: Google Play
- iOS: App Store (pending review)
Two models are actively trained and maintained:
| v3.5 — Mobile | v4 — Server | |
|---|---|---|
| Purpose | On-device / edge | Server-side / documents |
| Architecture | MobileNetV3 + 2×BiLSTM(512) + CTC | Swin-T Encoder + 6-layer Transformer Decoder |
| Parameters | 11.4M | ~54M |
| Input | Grayscale, 160px height |
RGB, 224×1024px |
| Export | ONNX FP32/FP16/INT8 · CoreML | ONNX only |
| Inference (CPU) | ~30ms/line | ~180ms/line |
The mobile model (v3.5) runs on-device across Web (WASM), Android (NNAPI), and iOS (Core ML). The server model (v4) handles complex document images with colour backgrounds and longer text sequences.
Because high-quality Mon datasets are scarce, validated samples from the app's feedback flow feed directly into future training rounds.
The mobile model (v3.5) deploys to Web, Android, and iOS — each using the format that enables hardware acceleration:
| Platform | Format | Acceleration |
|---|---|---|
| Web | ONNX | WASM |
| Android | ONNX | NNAPI |
| iOS | CoreML .mlpackage |
Apple Neural Engine |
- Web App — SvelteKit PWA
- Android App — Jetpack Compose
- iOS App — SwiftUI
- Feedback Service — Go ingestion API
- Shared Assets — model weights, locales, sync scripts
- HuggingFace — ONNX, CoreML, and checkpoint files
- npm package — JavaScript SDK
- Architecture decisions — ADRs
- API specs — OpenAPI contracts
- Mon Corpus Collection — training dataset
- Bugs: GitHub Issues
- Translations: Shared translation sheet
- Script samples: Contribute via the Android or iOS app, or reach out directly
- Standards: Contributing Guide · Security Policy