Fast KhmerOCR Offline Model
It is a fast Khmer + English OCR model that runs completely offline.
សូមចុច ដើម្បីជ្រើសរើស ឯកសារ
Drag & drop, paste (Ctrl+V), or select a file
Preview
លទ្ធផល
កំពុងដំណើរការ...
ប្រវត្តិការប្រើប្រាស់
បង្ហាញ 6 រូបភាពចុងក្រោយ
ចំនួនប្រើប្រាស់សរុប:
0
ឯកសារយោង
Text detection uses the existing PaddleOCR model, optimized to run on edge devices. Text recognition uses our in-house model trained from scratch (3.7 MB after ONNX conversion). Both models run completely offline with zero API calls or external services. The recognition model was trained with 16 million images across 600+ Khmer and English fonts.
The recognition model supports Khmer and English text and can process multi-line images end-to-end.
Available
- Offline detection + recognition (no API calls)
- Khmer + English support
- End-to-end, multi-line recognition
- Text detection 2.5 MB
- Text recognition 3.7 MB
Needs Improvement
- Emoji characters are not supported yet
- Long text lines (> 90 characters) are less accurate
- Stylized or artistic fonts remain challenging
- Curve Text and rotation text is not supported yet