X-LINUX-AI OpenSTLinux Expansion Package

Warning

Please read carefully

1. What is X-LINUX-AI?[edit source]

X-LINUX-AI is a free of charge open-source software package dedicated to AI.
It is a complete ecosystem that allow developers working with OpenSTLinux to create AI-based application very easily.

All-in-one AI solutions for the entire STM32MPU serie
Pre-integrated into Linux distribution based on ST environment
Include AI frameworks to execute Neural Network models
Include AI model benchmark application tools for MPU
Easy application prototyping using Python language and AI frameworks Python API
C++ API for embedded high-performance applications
Optimized open-source solutions provided with source codes that allow extensive code reuse and time savings
Delivered under GitHub as a Yocto meta-layer

2. X-LINUX-AI Software stack introduction[edit source]

X-LINUX-AI is an STM32 MPU OpenSTLinux Expansion Package that targets artificial intelligence for STM32MPx Series devices.
It contains Linux AI frameworks, development tools, as well as application examples to get started with some basic AI use cases such as computer vision.
It is composed of an OpenEmbedded meta layer, named meta-st-stm32mpu-ai, to be added on top of the STM32MPx Distribution Package. It brings a complete and coherent easy-to-build / install environment to take advantage of AI on STM32MPx Series devices.

Depending on the STM32MPU series X-LINUX-AI supports inference execution either on Neural Processing Unit (NPU), Graphics Processing Unit (GPU) or Cortex-A (with multi-threading).

X-LINUX-AI supported inference engine VS sSTM32MPUs series

2.1. Seamless integration of common AI frameworks[edit source]

2.2. AI framework details[edit source]

2.2.1. Tensorflow Lite runtime[edit source]

2.2.2. ONNX runtime[edit source]

3. X-LINUX-AI STM32MP25 ALPHA[edit source]

Information

This version is compatible with Yocto Project^® build system Kirkstone and has been validated against the OpenSTLinux STM32MP25 Ecosystem ALPHA V0.2.0 and validated on unknown board type identifier .

3.1. Contents[edit source]

TensorFlow™ Lite^[1] 2.11.0
Coral Edge TPU™^[2] accelerator native support
- libedgetpu 2.0.0 (Grouper) aligned with TensorFlow™ Lite 2.11.0
- libcoral 2.0.0 (Grouper) aligned with TensorFlow™ Lite 2.11.0
- PyCoral 2.0.0 (Grouper) aligned with TensorFlow™ Lite 2.11.0
ONNX Runtime™ ^[3] 1.11.0
OpenCV^[4] 4.5.x
Python™^[5] 3.10.x (enabling Pillow module)
Support of Sony™ IMX335 5Mpx sensor with use of DCMIPP and internal ISP
TensorFlow™ Lite application samples :
- NPU Python human pose estimation using TensorFlow™ Lite based on Movenet SinglePose Lightning quantized model
- NPU Python semantic segmentation using TensorFlow™ Lite based on DeepLabV3 quantized model
- NPU C++ / Python image classification using TensorFlow™ Lite based on MobileNet v1 quantized model
- NPU C++ / Python object detection using TensorFlow™ Lite based on COCO SSD MobileNet v1 quantized model
- NPU C++ face recognition using TensorFlow™ Lite models capable of recognizing the face of a known (enrolled) user (available on demand)

Warning

The face recognition binary is available on demand. Please contact the local STMicroelectronics support for more information about this application or send a request to edge.ai@st.com

Coral Edge TPU™ based application samples :
- C++ / Python image classification using Coral Edge TPU™ based on MobileNet v1 quantized model and compiled for the Coral Edge TPU™
- C++ / Python object detection using Coral Edge TPU™ based on COCO SSD MobileNet v1 quantized model and compiled for the Coral Edge TPU™
ONNX Runtime™ application samples :
- Python image classification using ONNX Runtime™ based on MobileNet v1 quantized model
- C++ / Python object detection using ONNX Runtime™ based on COCO SSD MobileNet v1 quantized model
OpenVX™ based application samples :
- NPU C++ image classification using Network Binary Graph based on COCO SSD MobileNet v1 quantized model
Tools :
- STM32AI MPU offline tool dedicated to deploy NN on the stm32mp25
- NBG benchmark application

- Support of Jupyterlab™ for python prototyping

3.2. Validated hardware[edit source]

X-LINUX-AI STM32MP25 beta is supported on all STM32MP25 Series and it has been validated on the following boards:

STM32MP257F-EV1 Evaluation board

4. X-LINUX-AI package[edit source]

X-LINUX-AI Starter package

Quickly and easily bring AI on STM32MPUs series

X-LINUX-AI Developer package

The X-LINUX-AI SDK add-on extends the OpenSTLinux SDK
with AI functionality to develop and build an AI application easily