Deploy PaddlePaddle on Arm Cortex-M with Arm Virtual Hardware

Try the end-to-end workflow

This section provides hands-on instructions for you to deploy pre-trained PaddlePaddle models on the Corstone-300 Fixed Virtual Platform (FVP) included with Arm Virtual Hardware.

The steps involved in the model deployment are shown in the figure below:

Image Alt Text:End-to-end workflowFigure 3. End-to-end workflow

Deploy PaddleOCR text recognition model on the Corstone-300 FVP included with Arm Virtual Hardware

Start by launching the Arm Virtual Hardware AMI .

Alternatively, you can also download the Corstone-300 FVP from the Arm Ecosystem FVP page. For installation instructions see Arm Ecosystem FVPs .

The code for this project , is available to download from ArmDeveloperEcosystem GitHub repository as well as PaddleOCR ’s GitHub repository (under the dygraph branch).

Start by cloning the code repository on your running AVH AMI instance:


            git clone
cd Paddle-examples-for-AVH/OCR-example/Text-recognition-example

In this directory, there is a script named that automates the entire process described in the End-to-end workflow diagram.

The script automatically builds and executes the English text recognition application on the Corstone-300 platform included with Arm Virtual Hardware. Here is a list of steps performed by this script:

  • Step 1. Set up build environment
  • Step 2. Download a trained PaddleOCR text recognition model
  • Step 3. Use TVMC to compile the model and generate code for the Arm Cortex-M processor
  • Step 4. Process resources for building the application image
  • Step 5. Build the target application using Makefiles
  • Step 6. Run application binary on Corstone-300 FVP included in AVH

Training the model usually takes a lot of time. In step 2, an already trained English text recognition model named ocr_en.tar is used.

By default, the script uses the image shown below (QBHOUSE) as an example to verify the inference results on the Corstone-300 FVP with Arm Cortex-M55.

Image Alt Text:QBHOUSE

You can now run the trained PaddleOCR text recognition model on the Corstone-300 FVP included on the AVH AMI with the following command:



The output from running the application on the Corstone-300 FVP is shown below:


        Ethos-U rev 136b7d75 --- Feb 16 2022 15:47:15
    (C) COPYRIGHT 2019-2022 Arm Limited

Starting ocr rec inference
text: QBHOUSE, score: 0.986746
Info: /OSCI/SystemC: Simulation stopped by user.


The text recognition results are consistent with the input image text content QBHOUSE and has a high confidence score of 0.986746.

You have successfully deployed a PP-OCRv3 English recognition model directly on the Corstone-300 FVP with the Arm Cortex-M55.