Movidius^TM神经计算棒的测评  

Evaluation of Movidius^TM Neural Compute Stick

作　　者：李倩 原建平[1] LI Qian;YUAN Jianping(Research Center for Ultrasonics and Technologies,Institute of Acoustics,Chinese Academy of Sciences,Beijing,100190,China;University of Chinese Academy of Sciences,Beijing,100049,China)

机构地区：[1]中国科学院声学研究所超声技术中心,北京100190 [2]中国科学院大学,北京100049

出　　处：《网络新媒体技术》2020年第2期51-59,共9页Network New Media Technology

摘　　要：Intel神经计算棒(NCS),世界上首个基于USB模式的深度学习推理工具和独立的人工智能(AI)加速器,作为最近流行的深度学习开发工具,适用于低功耗下的终端AI应用场景。针对NCS的高速计算优势,本文将NCS与CPU、GPU一起进行性能对比。首先,通过SqueezeNet模型来对比树莓派CPU、GPU、Pi+NCS三种工具的图片分类检测速度;其次,在同一张图片同时运行车辆类型和车牌检测2个模型,进一步进行PC CPU与NCS的性能对比。实验结果表明,NCS在低功耗的前提下具有高性能的运算能力,更适用于广泛的移动和嵌入式视觉设备。Intel Neural Compute Stick(NCS),the world’s first deep learning inference tool based on USB mode and independent artificial intelligence(AI)accelerator,as a popular deep learning development tool for low power terminals AI application scenario.For the high-speed computing advantages of NCS,this paper compares the performance of NCS with CPU and GPU.Firstly,compare the image classification detection speed of the Raspberry Pi CPU,GPU,Pi+NCS tools through the SqueezeNet model.Secondly,run the vehicle type and license plate detection two models simultaneously in the same picture to further carry out the PC CPU and NCS.Performance comparison.The experimental results show that NCS has high-performance computing power under the premise of low power consumption,and is more suitable for a wide range of mobile and embedded vision devices.

关 键 词：神经计算棒 深度学习 终端AI 高速计算 性能对比 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]