1. Developer package: build AI application from the X-LINUX-AI add-on SDK[edit source]
In order to easily develop and build an AI application without using the Yocto build system, X-LINUX-AI comes with a SDK add-on extending the OpenSTLinux SDK with AI functionality. To install and use the X-LINUX-AI SDK Add-on please refer to the following steps.
2. Prerequisites[edit source]
2.1. Install the OpenSTLinux SDK[edit source]
First of all, you must download and install the OpenSTLinux SDK, which contains all the basis needed for the X-LINUX-AI add-on. To do this, follow instructions given in STM32MPU Developer Package: Installing the SDK chapter. Once this has been done, you should have a directory containing the OpenSTLinux SDK.
Information |
Once this is done, if you have follow the example of directory structure the path to the OpenSTLinux SDK should be the following:
|
3. X-LINUX-AI SDK add-on installation[edit source]
3.1. Download the X-LINUX-AI SDK add-on[edit source]
To add the Artificial Intelligence part into the OpenSTLinux SDK, you must download and install the X-LINUX-AI SDK add-on. The add-on is delivered through a tarball file that can be downloaded here: X-LINUX-AI SDK add-on.
3.2. Install the X-LINUX-AI SDK add-on[edit source]
Host PC package prerequisites
To be able to properly install the X-LINUX-AI SDK add-on on your host computer it is necessary to install the extra package patchelf using the following command (in addition of packages from PC_prerequisites):
sudo apt-get install patchelf
Install add-on for STM32MP1 series
Once the X-LINUX-AI SDK add-on is downloaded, uncompress the add-on:
Information |
The name of the tarball depends on the chosen version. In the example below the last uploaded version is used. |
cd ~/Downloads/ tar xzf en.sdk-x86_64-stm32mp1-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0.tar.gz
Then, copy the .sh script to your OpenSTLinux SDK directory:
cp stm32mp1-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0/st-image-ai-openstlinux-weston-stm32mp1-x86_64-toolchain-4.2.4-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0.sh <working directory absolute path>/Developer-Package/SDK
Finally, run the script contained in the tarball:
cd <working directory absolute path>/Developer-Package/SDK
./st-image-ai-openstlinux-weston-stm32mp1-x86_64-toolchain-4.2.4-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0.sh
Optionally, once the script is executed, you can delete it.
rm *.sh
Install add-on for STM32MP25x lines
Once the X-LINUX-AI SDK add-on is downloaded, uncompress the add-on:
Information |
The name of the tarball depends on the chosen version. In the example below the last uploaded version is used. |
cd ~/Downloads/ tar xzf en.sdk-x86_64-stm32mp2-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0.tar.gz
Then, copy the .sh script to your OpenSTLinux SDK directory:
cp stm32mp2-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0/st-image-ai-openstlinux-weston-stm32mp2-x86_64-toolchain-4.2.4-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0.sh <working directory absolute path>/Developer-Package/SDK
Finally, run the script contained in the tarball:
cd <working directory absolute path>/Developer-Package/SDK
./st-image-ai-openstlinux-weston-stm32mp2-x86_64-toolchain-4.2.4-openstlinux-6.1-yocto-mickledore-mpu-v24.06.26-addon-x-linux-ai-v5.1.0.sh
Optionally, once the script is executed, you can delete it.
rm *.sh
3.3. Start the SDK[edit source]
Information |
The SDK environment setup script must be run once on each new working terminal on which you cross-compile. |
The add-on is now installed into the OpenSTLinux SDK. You can start the SDK. Go to your OpenSTLinux SDK directory and source the environment:
cd <working directory absolute path>/Developer-Package/SDK
source environment-setup-cortexa35-ostl-linux
export SYSROOT="<working directory absolute path>/Developer-Package/SDK/sysroots/cortexa35-ostl-linux"
source environment-setup-cortexa7t2hf-neon-vfpv4-ostl-linux-gnueabi
export SYSROOT="<working directory absolute path>/Developer-Package/SDK/sysroots/cortexa7t2hf-neon-vfpv4-ostl-linux-gnueabi"
Check that the SDK is properly installed:
x-linux-ai -v
X-LINUX-AI version: 5.1.0
Information |
With the x-linux-ai binary it is possible to check the X-LINUX-AI version and the supported frameworks/applications. Use x-linux-ai --help for more information. |
4. Use the SDK[edit source]
4.1. Build an application with the SDK[edit source]
Information |
In this section, the image classification application is used as an example, but all the applications can be built with this method. |
Once the SDK is correctly set up, the applications can be built easily. In this example, it is the image classification application that is built.
Download the github repository:
git clone https://github.com/STMicroelectronics/meta-st-x-linux-ai
Go to the image classification directory:
cd meta-st-x-linux-ai/recipes-samples/image-classification/files/stai-mpu
[Optional] Modify the application before compiling it. Add for example the following piece of code at the beginning of the main function:
Open the source code using vi or other code editor:
vi stai_mpu_image_classification.cc
Add a print function:
g_print("Application modified using OSTL SDK + X-LINUX-AI add-on \n");
Then, use the make command to build the application:
make
A new file is displayed, named stai_mpu_image_classification. This is a binary file, which has been generated using the make command. It is compiled for the architecture of the board corresponding to the chosen SDK. It must now be transferred to the board.
4.2. Use the application[edit source]
Once the board is correctly set up and the X-LINUX-AI packages are installed with the right version, it is possible to send the application to the board.
Information |
The X-LINUX-AI packages versions must be the same as the ones in the X-LINUX-AI SDK add-on. |
To do it, use the following command using your own IP address:
scp -r -p stai_mpu_image_classification root@<ip_address>:/usr/local/x-linux-ai/image-classification/
Then, use the ssh protocol to connect to the board:
ssh root@<ip_address>
On the board, go to the right directory and use the image classification script to launch the application:
cd /usr/local/x-linux-ai/image-classification/ ./launch_bin_image_classification.sh
This script is using the stai_mpu_image_classification application that has been compiled before.
[Optional] If in the previous part you have added the print in the main function, it should be visible in the logs of the application:
root@stm32mp2:/usr/local/x-linux-ai/image-classification# ./launch_bin_image_cla ssification.sh stai wrapper used : machine used = stm32mp257f-ev1st user : weston Application modified using OSTL SDK + X-LINUX-AI add-on model file set to: /usr/local/x-linux-ai/image-classification/models/mobilenet/mobilenet_v2_1.0_224_int8_per_tensor.nb label file set to: /usr/local/x-linux-ai/image-classification/models/mobilenet/labels_imagenet_2012.txt camera framerate set to: 30 camera frame width set to: 760 camera frame height set to: 568 dual DCMIPP camera pipeline mode enabled 2 cpu core(s) available Loading dynamically: /usr/lib/libstai_mpu_ovx.so.5 m_num_inputs 1 m_num_outputs 1 m_input_height 224 m_input_width 224 m_input_channels 3 m_sizeInBytes 150528 check if a CSI or USB camera is connected /home/weston dcmipp found display resolution is : 1024 x 600 Display config <= 600p video source used : /dev/video2 camera pipeline configuration : video/x-raw, format=RGB16, width=760, height=568, framerate=30/1,framerate=30/1 nn source used : /dev/video1 nn pipeline configuration : video/x-raw, format=RGB, width=224, height=224, framerate=30/1