AXERA-TECH/ax-samples

What it solves

AX-Samples provides a collection of example codes for implementing common open-source deep learning algorithms on Axera AI SoC (System on Chip) hardware. It simplifies the process for community developers to evaluate and adapt AI models to these specific hardware platforms.

How it works

The project provides a set of examples covering various deep learning tasks such as classification, detection, and pose estimation, as well as traditional computer vision operations. It utilizes a CMake-based cross-platform compilation system to allow developers to build the project for different Axera chips (including AX650N, AX637, AX630C, AX620Q, and AX615).

Who it’s for

Developers working with Axera AI SoC hardware, including those using development boards like AXera-Pi Pro, AI Pyramid, and LLM630 Compute Kit, who need to implement and run AI models on the edge.

Highlights

• Broad Hardware Support: Supports a wide range of Axera chips for terminal, automotive, and edge computing chips (e.g., AX650N, M57, AX8850N).
• Cross-Platform Compilation: Simple cross-platform compilation based on CMake.
• Ready-to-Use Resources: Provides links to ModelZoo, pre-compiled programs, and test images via Huggingface and Modelscope.
• Performance Benchmarks: Includes inference time statistics for common open-source models based on real-world testing on the AXera-Pi Pro board.