MaixPy

Features

• High-level Python 3 API for rapid prototyping of AI vision and audio projects.

• Integrated support for NPU-accelerated AI models including YOLOv5, YOLOv8, YOLO11, and MobileNet.

• Support for Large Language Models (LLM) and Vision Language Models (VLM) on edge devices.

• Unified API across Python and C++ (MaixCDK) for seamless transition from prototype to production.

• Hardware-accelerated video encoding and decoding (H.264/H.265/JPEG) up to 4K resolution.

• Comprehensive peripheral control for UART, I2C, SPI, and GPIO via standard Python modules.

• Built-in image processing algorithms compatible with OpenMV and OpenCV standards.

• Native support for multiple hardware platforms including MaixCAM, MaixCAM-Pro, and MaixCAM2.

• Dual-OS architecture support, running on Linux (Buildroot/Ubuntu) alongside real-time subsystems (RTOS/RTT).

• Integration with MaixHub for one-click online AI model training and deployment.

• MaixVision IDE support for visual debugging and code management.

• High-performance NPU utilization reaching up to 3.2 Tops@INT8 on supported hardware.

• Support for WiFi 6 and Bluetooth Low Energy (BLE 5.4) connectivity.

• Automated API documentation and stub generation using pybind11-stubgen.

• Extensible component-based architecture allowing for custom C++ module integration.

MaixPy v4 is built as a multi-layered ecosystem that bridges high-level scripting with low-level hardware acceleration. At its core, the library acts as a Python wrapper for MaixCDK (C++ Development Kit), utilizing pybind11 to expose high-performance C++ components to the Python environment. This design allows developers to leverage the ease of Python for application logic while maintaining the execution speed of C++ for compute-intensive tasks like NPU inference and image processing.

The system architecture is designed to be cross-platform, supporting various backends including Linux (Buildroot or Ubuntu) and real-time operating systems (RTOS/RTT). It employs a component-based structure where modules for camera input, display output, neural network inference (NN), and hardware peripherals are decoupled, allowing for flexible project configurations.

Core Components:

• Maix Module: The primary Python namespace containing sub-modules for camera, display, image, nn, and uart.
• MaixCDK: The underlying C++ SDK providing the implementation for hardware abstraction and NPU drivers.
• NPU Accelerator: Specialized logic for executing quantized INT8 models (YOLO, ResNet, etc.) with hardware acceleration.
• Multimedia Pipeline: Handles high-speed video streaming, encoding (H.264/H.265), and decoding.

Use Cases

This library is ideal for:

• Edge AI Deployment: Rapidly deploying object detection, image classification, or face recognition models to low-power embedded devices using Python.
• STEM Education: Teaching AI and robotics concepts through an accessible Python API that abstracts complex hardware and driver interactions.
• Industrial Quality Assurance: Building automated visual inspection systems for production lines that require real-time processing and peripheral control (e.g., triggering actuators via UART/GPIO).
• Smart IoT Devices: Developing connected devices that require vision-based intelligence, such as smart security cameras or interactive kiosks with LLM capabilities.

Getting Started

To begin developing with MaixPy, it is recommended to use a supported hardware platform like MaixCAM. Developers can install the maixpy package via pip on compatible Linux environments or use the pre-installed firmware on Sipeed hardware. For those wishing to build from source, the process involves setting up the MaixCDK environment variable (MAIXCDK_PATH) and running the project.py build script. Detailed documentation, including API references and quick-start tutorials, is available at the official Sipeed Wiki. For a visual development experience, the MaixVision IDE provides a workstation-like environment for coding and real-time debugging.