Performance analysis of real-time object detection on Jetson device
Jiawei Zhu, Haogang Feng, Shida Zhong, Tao Yuan
Abstract
As the development of 5G communications and Internet of things, more and more devices are required to run with low-power consumption and high performance as edge devices, and most of them have to do edge computing such as running the inference part of the artificial intelligence (AI) model to handle intelligent applications. In this paper, we deploy YOLOv3 and PPYOLO models on two kinds of intelligence edges, which are Jetson Nano and Jetson Xavier NX, to benchmark their performance. The tiny versions of YOLOv3 and PPYOLO are further deployed on Jetson Nano with relatively limited computing resources to make a comparative analysis. The benchmark results are analyzed and investigated to give empirical recommendation for reference of developing and deploying intelligent real-time object detection applications on intelligence edges to meet different requirements.