MaixPy (v4) simplifies the creation of AI projects on edge devices using Python. It targets makers, students, and engineers seeking to rapidly prototype and deploy AI vision, audio, and LLM/VLM applications with minimal AI expertise. The project offers an integrated hardware-software ecosystem, including an IDE, online training platform, and C/C++ SDKs, enabling seamless development from prototype to production.

How It Works

MaixPy provides a Pythonic interface for interacting with edge hardware peripherals, camera streams, and AI models. It leverages specialized hardware platforms like MaixCAM and MaixCAM2, which feature dedicated NPUs and improved processing capabilities compared to older generations. The framework supports various AI tasks, including image classification and object detection, with optimized models and efficient data flow. An accompanying ecosystem, including MaixHub for online AI model training and MaixVision IDE, further streamlines the development workflow.

Quick Start & Requirements

• Installation: pip install maixpy
• Prerequisites: Specific hardware platforms (MaixCAM, MaixCAM2) are recommended for optimal performance. Python 3 is required. Links to Quick Start, Documentation, and API are available.
• Hardware: Features powerful RISC-V and ARM CPUs, significant memory (up to 4GB LPDDR4), and dedicated NPUs (up to 3.2Tops@INT8). Supports WiFi6, BLE5.4, and various display/camera interfaces.

Highlighted Details

• Significant performance improvements over older K210 hardware, with MaixCAM2 achieving up to 1218fps for MobileNetv2 classification and 495fps for YOLOv5s detection.
• Support for Large Language Models (LLMs) and Vision-Language Models (VLMs) on MaixCAM2, including models like Qwen/DeepSeek and InterVL.
• MaixHub provides a free online platform for training AI models with a single click, abstracting complex AI knowledge.
• Offers a C/C++ SDK with identical APIs to the Python SDK, facilitating smooth transitions to production environments.

Maintenance & Community

The project is actively developed by Sipeed Ltd. Community engagement is fostered through MaixHub (sharing, discussion), QQ groups (862340358), Telegram (t.me/maixpy), and GitHub issues.

Licensing & Compatibility

All files in this repository are under the terms of the Apache License 2.0, with exceptions for third-party libraries that may have their own licenses. This license permits commercial use and integration into closed-source projects.

Limitations & Caveats

Performance metrics vary significantly based on the specific MaixCAM hardware model, resolution, and AI model used. While v4 supports LLMs/VLMs, this capability is primarily highlighted for the newer MaixCAM2 hardware. Older hardware like K210 is noted as having many limitations in memory and performance.