Note

This is the documentation for the latest development branch and may refer to features that are not available in released versions. If you are looking for the documentation for a specific release, use the drop-down menu on the left and select the desired version.

AI Application Development Guide

This chapter focuses on AI scenario development, covering everything from basic inference calls to multi-task, multi-channel input, and complete scenario-based examples.

Detailed Guide

• NNCASE Application Guide
• AI2D Application Development Guide
• KPU Application Development Guide
• Single-Model Application Development Guide
• Dual-Model Application Development Guide
• Triple-Camera AI Application Development Guide
• UVC+AI Application Development Guide
• YOLO Application Guide
• RT-Smart Deployment of Kanzhi Online Training Platform Model
• AI Demo Application Guide
• Multi-Object Tracking (MOT) Application Development Guide

The AI application examples for the RT-Smart SDK are located in the src/rtsmart/examples/ai directory. To help you progress from “running a demo” to “independent development,” we have divided the examples into the following three major categories:

Core Acceleration Introduction (Basics)

Objective: Understand K230 hardware acceleration implementation If you are new to the Canaan K230 chip, it is recommended that you study these examples first to master the fundamentals of hardware-accelerated computing.

Application Directory	Function Description	Core Value
usage_ai2d	Demonstrates the 5 types of preprocessing supported by hardware (cropping, scaling, padding, affine transformation, shifting).	Master how to use the AI2D hardware to offload CPU pressure.
usage_kpu	Using YOLOv8 as an example, this fully demonstrates the complete workflow from model loading, preprocessing, inference, to post-processing.	Learn the KPU low-level API call logic and Tensor processing.

Classic Task Templates (Advanced)

Objective: Quickly build applications based on the encapsulated framework These examples provide standardized code structures, suitable for quickly familiarizing yourself with AI application development logic and building AI applications based on the encapsulated framework by following the given examples.

Application Directory	Task Type	Applicable Scenario
face_detection	Single-model task	Develop the most basic single-point AI functionality.
face_recognition	Multi-model cascade	Learn the pipeline logic of detection + recognition.
triple_camera_ai	Multi-camera vision	Demonstrate the architecture of three-channel camera input simultaneously running AI inference.
uvc_face_detection	UVC input	Learn how to drive a USB camera (UVC) for AI analysis.
yolo	General-purpose encapsulated tool	Highly encapsulated, supports YOLOv5/v8/v11, covering classification, detection, segmentation, and rotated detection.
cloudplat_deploy	Platform model deployment	Quickly validate models obtained from the online training platform or AICube.

Scenario-based Applications (Practical)

Objective: Reference for product-level development solution prototypes These examples provide deep encapsulation of underlying multimedia (ISP camera, VO display, video encoding/decoding), primarily demonstrating K230’s scenario adaptation capabilities.

Application Directory	Function Description	Applicable Scenario
ai_demo	Integrates 50+ scenario-based examples, covering object recognition, face detection, gesture recognition, human body recognition, license plate recognition, OCR text recognition, etc.	Evaluate K230’s performance upper limit and quickly find business prototypes.
multi_object_tracking	Integrates various commonly used multi-object tracking (MOT) algorithms.	Suitable for dynamic analysis scenarios such as security surveillance and people flow statistics.