PP-OCRv4 Mobile Det
PaddlePaddle/PP-OCRv4_mobile_det
published Jun 2025 · updated Jul 2025
PP-OCRv4 Mobile Det is an image-to-text model that detects text regions in images, outputting bounding polygons with confidence scores, optimized for mobile and edge devices.
specs
| Task | Text Detection |
| Architecture | Mobile-optimized convolutional neural network |
| License | Apache 2.0 |
| Accuracy (Average) | 0.624 |
| Output Format | Polygons and confidence scores |
about this model
PP-OCRv4_mobile_det is a text detection model that localizes text regions in images, optimized for mobile and edge deployment. Developed by the PaddleOCR team as part of the PP-OCRv4_det series, it balances high accuracy with efficient inference, making it suitable for resource-constrained environments. The model outputs polygon coordinates and confidence scores for detected text boxes.
Benchmark Performance
The model's accuracy (measured as F-score) across diverse text scenarios is as follows:
| Scenario | H-Chinese | H-English | P-Chinese | P-English | T-Chinese | Ancient | Japanese | General | Pinyin | Rotation | Distortion | Artistic | Average |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| F-score | 0.583 | 0.369 | 0.872 | 0.773 | 0.663 | 0.231 | 0.634 | 0.710 | 0.430 | 0.299 | 0.715 | 0.549 | 0.624 |
H = Handwritten, P = Printed, T = Traditional. Average F-score across all scenarios is 0.624.
Output Example
The following image shows a typical detection result with detected polygons overlaid on the input:
PP-OCRv4_mobile_det is licensed under Apache 2.0 and is available as a hosted API on gigarouter, requiring no local installation. It integrates seamlessly with the PP-OCRv4 recognition pipeline for end-to-end OCR workflows.
best for
- ·Real-time OCR on mobile devices
- ·Document scanning and digitization
- ·Extracting text from images on edge devices
FAQ
It detects text regions in images, typically used as part of an OCR pipeline to extract text.
Apache 2.0.
Input: image (URL or file). Output: polygon coordinates and confidence scores for each detected text region.
Use the gigarouter OpenAI-compatible endpoint with your API key, specifying the model name "PP-OCRv4 Mobile Det".
Yes, it is mobile-optimized for efficient inference on edge devices.
We're benchmarking and onboarding PP-OCRv4 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.