ALLPCB
Introduction
This article shows how to validate MobileNet image classification inference on the iTOP-RK3568 AI development board.
Board hardware and capabilities
The board is based on the Rockchip RK3568 processor, with a quad-core Cortex-A55 cluster and a Mali-G52 GPU. It integrates an NPU with up to 1 TOPS of peak compute, which is suitable for lightweight mobile models such as MobileNet. The board provides multiple hardware interfaces, supports direct connection of MI PI camera modules for image capture, and allows data import via USB devices. A development kit is available with model conversion tools and inference examples to assist model deployment and testing.
Example: MobileNet inference test
Copy the files in that directory to the development board and extract them. The extraction result should look like this:
Enter the folder and run the executable to perform image inference using the following command:
export LD_LIBRARY_PATH=./lib./rknn_mobilenet_demo model/mobilenet_v2.rknn model/bell.jpg
The program prints the inference result for the input image. In this example, the highest-confidence label corresponds to the bell class, and the output confirms successful model inference:
Notes
The article content and images were provided by the contributing author or a partner site. The views are those of the author and do not represent the site's position. Materials are provided for engineers' learning purposes only.
