作者:Andrew G. Howard,Menglong Zhu,Bo Chen,Dmitry Kalenichenko,Weijun Wang,Tobias Weyand,Marco Andreetto,Hartwig Adam

作者单位:Google

发布时间:2017

发布期刊/会议:CVPR

论文全称:MobileNets:Efficient convolutional neural networks for mobile vision applications

论文地址:https://arxiv.org/abs/1704.04861

论文代码:

地位:针对移动端计算机视觉应用的高效卷积神经网络

个人理解

一、摘要

We present a class of effificient models called MobileNets for mobile and embedded vision applications. MobileNets are based on a streamlined architecture that uses depthwise separable convolutions to build light weight deep neural networks. We introduce two simple global hyper parameters that effificiently trade off between latency and accuracy. These hyper-parameters allow the model builder to choose the right sized model for their application based on the constraints of the problem. We present extensive experiments on resource and accuracy tradeoffs and show strong performance compared to other popular models on ImageNet classifification. We then demonstrate the effectiveness of MobileNets across a wide range of applications and use cases including object detection, fifinegrain classifification, face attributes and large scale geo-localization.