PP-OCRv4 Mobile Rec (English)
PaddlePaddle/en_PP-OCRv4_mobile_rec
published Jun 2025 · updated Jul 2025
PP-OCRv4 Mobile Rec (English) is an image-to-text model that performs English text line recognition from images, optimized for mobile and edge devices.
specs
| Task | Image-to-Text (Text Recognition) |
| Architecture | PP-OCRv4 mobile recognition |
| Storage Size | 6.8 MB |
| Accuracy | 70.39% average accuracy |
| License | Apache 2.0 |
about this model
en_PP-OCRv4_mobile_rec is an image-to-text model that recognizes English and numeric characters from text line images. It is an English-specific variant of the PP-OCRv4_mobile_rec series developed by the PaddleOCR team.
Key Strengths
The model is ultra-lightweight with a storage size of 6.8 MB, making it suitable for edge deployment. It achieves a recognition average accuracy of 70.39% on standard benchmarks, with the evaluation metric requiring every character (including punctuation) in a line to be correct for the line to count as correct. This strict metric ensures high reliability in practical applications.
Accuracy and Specifications
| Model | Recognition Avg Accuracy (%) | Model Storage Size (M) | Introduction |
|---|---|---|---|
| en_PP-OCRv4_mobile_rec | 70.39 | 6.8 M | An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and numeric character recognition. |
Example Output
The following image shows the model’s recognition result on a sample text line:
Additional Details
- Licensed under Apache 2.0.
- Supports Python 3.8 through 3.12.
- Runs on CPU, GPU, XPU, and NPU hardware, and on Linux, Windows, and macOS operating systems.
- The PaddleOCR ecosystem has 70k+ GitHub stars and is used by 6k+ repositories, including integrations like Dify and RAGFlow.
For further information, refer to the PaddleOCR repository and the documentation.
best for
- ·English text extraction from scanned documents
- ·Real-time OCR on mobile or edge devices
- ·As a lightweight text recognition module in an OCR pipeline
FAQ
The model achieves 70.39% average accuracy on English text line recognition, where an entire line is marked incorrect if any character or punctuation is wrong.
The model is about 6.8 MB, making it suitable for mobile and edge deployment.
It supports English character and numeric recognition.
The model is licensed under Apache 2.0, as part of the PaddleOCR project.
Use the gigarouter OpenAI-compatible endpoint with your API key, specifying the model name en_PP-OCRv4_mobile_rec.
We're benchmarking and onboarding PP-OCRv4 Mobile Rec (English) 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.