This project provides sample code for implementing common deep learning algorithms on AXera's AI System-on-Chips (SoCs), targeting developers working with computer vision applications. It facilitates rapid evaluation and adaptation of AI models on specific hardware, benefiting embedded systems engineers and researchers needing to leverage specialized AI acceleration.

How It Works

The project implements popular open-source deep learning algorithms and traditional computer vision operations as example codebases. These are designed to run on AXera's proprietary AI SoCs, leveraging their NPU (Neural Processing Unit) capabilities. The approach focuses on providing practical, runnable examples that demonstrate the hardware's potential for efficient on-device inference.

Quick Start & Requirements

• Build: Compilation is managed via CMake, supporting cross-platform builds.
• Prerequisites: Requires specific AXera AI SoCs (e.g., AX650N, AX637, AX630C, AX620Q, AX615) and associated development boards like the AXera-Pi Pro. Access to NPU toolchains (Pulsar2) is necessary.
• Resources: Pre-compiled programs, test images, and model weights are available via Huggingface and Modelscope.
• Links:
  • NPU Toolchain Docs: [Link not directly provided, refer to Pulsar2 docs]
  • Pulsar2 Toolchain & Docs: [Link not directly provided, search for Pulsar2 download]
  • Development Board Wiki/Docs: Linked per board (e.g., AXera-Pi Pro Wiki).

Highlighted Details

• Supports a range of AXera terminal, edge, and automotive AI chips including AX650A/N, AX637, AX630C/AX620Q, AX615, and M57.
• Provides example code for common tasks like classification, detection, and pose estimation.
• Includes benchmark results for common open-source model inference times on the AXera-Pi Pro.
• Offers development board support through partners like Sipeed and M5Stack.

Maintenance & Community

• Code examples are continuously updated based on GitHub issues.
• Community support is available via GitHub Issues and a QQ group (139953715).
• Associated toolchain documentation (Pulsar2) is available but noted as potentially not up-to-date.

Licensing & Compatibility

• No specific open-source license is mentioned in the provided text.
• Compatibility is strictly limited to AXera AI SoC hardware. Commercial use implications are unclear without a stated license.

Limitations & Caveats

• The project is tightly coupled to proprietary AXera hardware, limiting its applicability outside that ecosystem.
• Toolchain documentation availability and update frequency may pose challenges for users.
• The absence of a clear license requires clarification for any form of adoption or distribution.