1. Introduction#
The K230 chip is the latest generation SoC product in Canaan Technology’s Kendryte® series of AIoT chips. This chip adopts a new multi-heterogeneous unit accelerated computing architecture, integrating two high-efficiency RISC-V computing cores and a new generation KPU (Knowledge Process Unit) intelligent computing unit. It features multi-precision AI computing power, widely supports general AI computing frameworks, and achieves utilization rates exceeding 70% for some typical networks.
The chip also provides a variety of peripheral interfaces and includes multiple specialized hardware acceleration units for scalar, vector, and graphic computations, such as 2D and 2.5D units. These enable full-process computation acceleration for diverse tasks, including image, video, audio, and AI tasks, ensuring low latency, high performance, low power consumption, quick startup, and high security.
AI development on the CanMV-K230 development board involves steps such as data preparation, model training, model conversion, KModel generation, image burning, and board operation. This document summarizes the resources and tools for AI development using the CanMV-K230, providing solutions for different scenarios. It also offers detailed operational tutorials for corresponding tools and AI application development code tutorials to improve user development efficiency.
For different users, we recommend focusing on different chapters as needed and progressing step by step.
Chapter |
Key Points |
Learning Guide |
Type |
Difficulty |
Adapting image |
|---|---|---|---|---|---|
Introduces K230_SDK and nncase, forming the basis for environment setup and model conversion on the K230 development board. |
Foundation for AI application development on the K230 development board. |
Basic |
I |
- Linux+RT-Smart |
|
Introduces over 50 AI Demo examples for different scenarios, providing source code addresses and operational tutorials. |
Suitable for users interested in running AI Demos on K230 and developing Demos for other scenarios. |
Application |
II |
- Linux+RT-Smart |
|
Introduces POCs implemented using the K230 development board, providing source code addresses and operational tutorials. |
Suitable for users interested in POCs supported by K230 and building practical application verifications. |
Application |
II |
- Linux+RT-Smart |
|
Introduces multimedia-related content in the K230 development process, including video frame acquisition and screen display, with C++ and MicroPython source code tutorials. |
Suitable for users interested in the AI inference process on K230, unfamiliar with multimedia, and capable of writing and debugging code independently. |
Tutorial |
III |
- Linux+RT-Smart |
|
Uses face detection and recognition as examples to detail the process of training, model conversion, development board code writing, and board operation using third-party open-source projects. |
Suitable for users familiar with the K230 development board and needing to run other open-source models on it, capable of writing and debugging code independently. |
Tutorial |
IV |
- Linux+RT-Smart |
|
Introduces various development tools provided by Canaan, explaining their advantages, limitations, and applicable user scenarios. |
Suitable for users interested in K230 development tools. |
Basic |
I |
- Linux+RT-Smart |
|
The online cloud training platform is a tool for model training using remote computing resources and a graphical interface. It introduces how to complete image classification and object detection tasks using the platform. |
Suitable for users interested in training models using Canaan’s online cloud training platform and deploying them on the board. |
Tool |
II |
- Linux+RT-Smart |
|
AI Cube is a tool for model training using local computing resources and a graphical interface. It introduces how to complete tasks on Ubuntu and Windows platforms and deploy them on the K230 development board. |
Suitable for users interested in training models using Canaan’s AI Cube and deploying them on the board. |
Tool |
II |
- Linux+RT-Smart |
|
Uses KTS (k230_training_scripts) to complete image classification and object detection tasks, providing training and deployment source code, and explaining each step from the source code perspective. |
Suitable for users interested in application development from the source code level, deploying on the board, and understanding each step from the code level. |
Tool, Tutorial |
III |
- Linux+RT-Smart |
|
Introduces the CanCollectorPlus data collection tool for the development board, providing a quick start tutorial. |
Suitable for users interested in the CanCollectorPlus tool or facing inconsistencies in training and deployment results due to dataset issues. |
Tool |
II |
- Linux+RT-Smart |