PP-OCRv5 Mobile Rec
PaddlePaddle/PP-OCRv5_mobile_rec
published Jun 2025 · updated Jul 2025
PP-OCRv5 Mobile Rec is an image-to-text model that recognizes text lines in multiple languages including Simplified Chinese, Traditional Chinese, English, and Japanese, as well as handwriting, vertical text, pinyin, and rare characters.
specs
| Task | Image-to-text (text line recognition) |
| Framework | PaddlePaddle |
| Language Support | Simplified Chinese, Traditional Chinese, English, Japanese, plus handwriting, vertical text, pinyin, rare characters |
| Average Accuracy | 0.8015 (line-level accuracy) |
about this model
Key Strengths
The model delivers strong accuracy across a diverse set of challenging conditions. Its performance is measured using a strict line-level metric where a single character error marks the entire line as incorrect, ensuring high practical reliability. Key accuracy results include:
| Scenario | Accuracy |
|---|---|
| Printed Chinese | 0.8605 |
| Printed English | 0.8753 |
| Traditional Chinese | 0.7199 |
| Japanese | 0.7577 |
| Handwritten Chinese | 0.4166 |
| Handwritten English | 0.4944 |
| Ancient Text | 0.5786 |
| Artistic Text | 0.5398 |
| General Scenario | 0.5570 |
| Pinyin | 0.7703 |
| Rotation | 0.7248 |
| Distortion | 0.8089 |
| Average | 0.8015 |
This model is part of the PP-OCRv5 series, which builds on earlier generations (PP-OCRv2, PP-OCRv3) that demonstrated improvements of 5% on Chinese scene text and 11% on English scene text over prior versions. The mobile variant is optimized for efficient deployment while maintaining high recognition quality.
Example output from the model demonstrates reliable transcription, such as recognizing the text "day as a reminder of the" with a confidence score of 0.979.
When used as part of the full PP-OCRv5 pipeline, which includes optional document orientation classification, text unwarping, and text detection modules, the model can handle complete OCR workflows on complex documents, as shown in the example below.
best for
- ·Recognizing printed or handwritten text in images for document digitization
- ·Multilingual OCR in applications supporting Chinese, English, and Japanese text
- ·Extracting text from complex scenarios like vertical text, artistic text, or distorted text
FAQ
It supports Simplified Chinese, Traditional Chinese, English, and Japanese, as well as handwriting, vertical text, pinyin, and rare characters.
The line-level accuracy average is 0.8015 across 12 scenarios including handwritten, printed, traditional, and artistic text.
Send an image to the gigarouter OpenAI-compatible endpoint with your API key. The model returns the recognized text in JSON format.
Yes, the model is designed to work within the PP-OCR pipeline, combining with text detection and orientation modules for end-to-end OCR.
We're benchmarking and onboarding PP-OCRv5 Mobile Rec 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.