Product overview
Key Benefits
Run AI models on STM32MPUs
Contains Linux® AI frameworks, as well as application examples.
XNNPACK support for TensorFlow™ Lite and ONNX Runtime
Convert and optimize deep neural networks using the tool.
Part of the ST Edge AI Suite
A collection of free online tools, case studies, and resources to support engineers at every stage of their edge AI development.
Description
X-LINUX-AI is an STM32 MPU OpenSTLinux Expansion Package that targets artificial intelligence for STM32MP1 and STM32MP2 series microprocessors. It contains Linux® AI frameworks, as well as application examples to get started with some basic use cases.
The examples provided in X-LINUX-AI include a selection of optimized models for image classification, object detection, semantic segmentation, human pose estimation, and face recognition.
These examples rely on the STAI_MPU API based on either the TensorFlow™ Lite inference engine, ONNX Runtime, or OpenVX™. They all support Python™ scripting and C/C++ applications.
-
All features
- Software environment
- The X-LINUX-AI OpenSTLinux Expansion Package v6.1.0 is compatible with the Yocto Project® build system Scarthgap. It is validated over the OpenSTLinux Distribution v6.1
- AI frameworks
- STAI_MPU unified API based on OpenVX™ (STM32MP23xx and STM32MP25xx only), TensorFlow™ Lite, and ONNX Runtime compatible with all STM32 MPU series
- TIM-VX™ 1.2.22 (STM32MP23xx and STM32MP25xx only)
- ONNX Runtime 1.19.2 with XNNPACK execution engine activated for CPU execution and VSINPU execution provider to address STM32MP2 NPU
- TensorFlow™ Lite 2.18.0 with VX-delegate (NPU/GPU for STM32MP25xx and STM32MP23xx only) and XNNPACK delegate activated (CPU only)
- ONNX 1.16.2 Python™ version for on-device learning
- PyTorch™ 2.3.1 Python™ version for on-device learning
- Applications
- Image classification
- C++ / Python™ example using STAI_MPU unified API based on the MobileNet v1 and v2 quantized models
- Object detection
- C++ / Python™ example using STAI_MPU unified API based on the SSD MobileNet v1 and v2 quantized models
- Face recognition
- C++ example using STAI_MPU unified API based on the BlazeFace and FaceNet quantized models (STM32MP23xx and STM32MP25xx only)
- Human pose estimation (STM32MP23xx and STM32MP25xx only)
- Python™ example using STAI_MPU unified API based on YOLOv8n pose quantized model
- Semantic segmentation (STM32MP23xx and STM32MP25xx only)
- Python™ example using STAI_MPU unified API based on DeepLabv3 quantized model
- People tracking and heatmap (STM32MP23xx and STM32MP25xx only)
- Python™ example using STAI_MPU unified API based on the YOLOv8n quantized models
- On-device learning for object detection (STM32MP23xx and STM32MP25xx only)
- Python™ example using STAI_MPU unified API based on SSD MobileNet v2 as the student and RT-DETR model as the teacher
- Step-by-step JupyterLab notebook available for this application
- Application examples based on Gstreamer 1.22.x, GTK® 3.x, OpenCV 4.9.x, Pillow, Python™ 3
- Image classification
- On-target utilities
- X-LINUX-AI tool suite: software information, management of AI packages, and benchmarking of neural network models
- Support for a wide range of image sensors and camera modules for STM32 MPUs including
- IMX335 (5MP) for STM32MP2 with the use of its internal ISP
- GC2145
- OV5640 for STM32MP13xx
- Host tools
- X-LINUX-AI SDK add-on extending the OpenSTLinux SDK with AI functionality to develop and build an AI application easily. The X-LINUX-AI SDK add-on supports all the above frameworks. It is available from the X-LINUX-AI product page
- Optimization and deployment of AI with the offline compiler available through the ST Edge AI Core (STEdgeAI-Core)
- Software environment