PP-OCRv5 Mobile Det
PaddlePaddle/PP-OCRv5_mobile_det
published Jun 2025 · updated Jul 2025
PP-OCRv5 Mobile Det is a text detection model that identifies and localizes text in images, supporting handwritten, printed, rotated, curved, and multi-language text including Chinese, English, and Japanese.
specs
| Task | Text Detection |
| License | Apache 2.0 |
| Supported Languages | Simplified Chinese, Traditional Chinese, English, Japanese |
| Average Detection Accuracy | 0.770 (over 13 scenarios) |
about this model
PP-OCRv5_mobile_det is a text detection model that extracts text regions from images, supporting multiple languages and challenging scenarios including handwriting, vertical, rotated, and curved text. It is part of the PP-OCRv5 series developed by the PaddleOCR team.
Key Strengths
- Robust handling of complex layouts, varying text sizes, and challenging backgrounds.
- Supports Simplified Chinese, Traditional Chinese, English, Japanese, and more.
- Licensed under Apache 2.0; compatible with Python 3.8–3.12 and hardware including CPU, GPU, XPU, and NPU.
- The PaddleOCR project has over 70,000 GitHub stars and is used by 6,000+ repositories.
Performance Benchmarks
| Handwritten Chinese | Handwritten English | Printed Chinese | Printed English | Traditional Chinese | Ancient Text | Japanese | General Scenario | Pinyin | Rotation | Distortion | Artistic Text | Average |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.744 | 0.777 | 0.905 | 0.910 | 0.823 | 0.581 | 0.727 | 0.721 | 0.575 | 0.647 | 0.827 | 0.525 | 0.770 |
Hosted on gigarouter as a managed API, the model requires no local installation — simply call the endpoint with an image to receive detected text polygons and confidence scores.
best for
- ·Document text detection in scanned documents and PDFs
- ·License plate detection from vehicle images
- ·Multi-language scene text detection in natural environments
FAQ
It accepts images and outputs bounding polygons (dt_polys) and confidence scores (dt_scores).
Simplified Chinese, Traditional Chinese, English, and Japanese.
Yes, it is designed to detect rotated, curved, and distorted text.
Apache 2.0.
Use the OpenAI-compatible endpoint with your API key; details are in the gigarouter documentation.
We're benchmarking and onboarding PP-OCRv5 Mobile Det as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.