邢培银,刘晓非,田永鸿

• 1:北京大学计算机科学技术系
• 2:云南大学信息学院
• 3:鹏城实验室

摘要(Abstract)：

数字视网膜架构是解决智慧城市中视频难汇聚、模型更新难问题的有效手段,其采用"特征实时汇聚、视频按需调取、模型动态更新"三种计算机制,解决传统以云服务器为中心的集中式视频处理框架存在的诸多问题。为了减少汇聚到云服务器的数据量,数字视网膜采用端边云协同计算架构,其中深度神经网络模型的计算被划分成两个部分,实现中低层特征提取的前一部分计算在前端设备或边服务器完成,再将提取的视觉特征压缩到极小数据量传送到云服务器,并在其上执行后一部分计算以实现高层特征分析与识别。由于系统中不同设备类型的计算能力各异,如何对不同时刻的视频分析处理任务合理分配计算资源,以达到整个系统的性效最优,是一个重要的研究问题。为此,本文对数字视网膜架构中深度神经网络模型的计算分配问题进行了形式化建模,提出了基于边云状态的启发式深度网络模型动态划分方法,并进一步针对云计算中心的异构设备提出了一种启发式计算任务分配算法。实验结果表明,与传统以云服务器为中心的集中式视频处理架构相比,本文所提方法可显著降低视频大数据处理系统的整体延迟。

关键词(KeyWords)： 端边云协同;数字视网膜;深度神经网络;动态划分

Abstract:

Keywords:

基金项目(Foundation):

作者(Author): 邢培银,刘晓非,田永鸿

DOI: 10.16453/j.cnki.ISSN2096-5036.2021.05.012

参考文献(References)：

• [1]Olga Russakovsky,Jia Deng,Hao Su et al.Imagenet large scale visual recognition challenge[J].International Journal of Computer Vision,2015,115(3):211-252.
• [2]Ross Girshick,Jeff Donahue,Trevor Darrell et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2014:580-587.
• [3]Alex Graves,Abdel-rahman Mohamed and Geoffrey Hinton.Speech recognition with deep recurrent neural networks[C]//Proceedings of the IEEE International Conference on Acoustics,Speech and Signal Processing.IEEE,2013:6645-6649.
• [4]Karen Simonyan and Andrew Zisserman.Two-stream convolutional networks for action recognition in videos[C]//Proceedings of the International Conference on Neural Information Processing Systems,2014.
• [5]Alessandro Redondi,Luca Baroffio,Lucio Bianchi et al.Compress-then-analyze versus analyze-then-compress:What is best in visual sensor networks?[J].IEEE Transactions on Mobile Computing,2016,15(12):3000-3013.
• [6]Wen Gao,Siwei Ma,Lingyu Duan,Yonghong Tian,Peiyin Xing,Yaowei Wang,Shanshe Wang,Huizhu Jia,Tiejun Huang.Digital Retina:A Way to Make the City Brain More Efficient by Visual Coding[J].IEEE Transactions on Circuits and Systems for Video Technology,2021.
• [7]Hyomin Choi and Ivan V Baji.Deep feature compression for collaborative object detection[C]//Proceedings of the IEEEInternational Conference on Image Processing.IEEE,2018:3743-3747.
• [8]Hyomin Choi and Ivan V Baji.Near-lossless deep feature compression for collaborative intelli-gence[C]//Proceedings of the IEEE International Workshop on Multimedia Signal Processing.2018:1-6.
• [9]Amir Erfan Eshratifar,Mohammad Saeed Abrishami and Massoud Pedram.Joint DNN:an eflcient training and inference engine for intelligent mobile cloud computing services[J].IEEE Transactions on Mobile Computing,2021,20:565-576.
• [10]Jongwon Choi,Hyung Jin Chang,Tobias Fischer et al.Context-aware deep feature compression for high-speed visual tracking[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2018:479-488.
• [11]Lin Ding,Yonghong Tian,Hongfei Fan et al.Rate-performance-loss optimization for inter-frame deep feature coding from videos[J].IEEE Transactions on Image Processing,2017,26(12):5743-5757.
• [12]Zhuo Chen,Kui Fan,Shiqi Wang et al.Intermediate deep feature compression:Toward intelligent sensing[J].IEEETransactions on Image Processing,2020,29:2230-2243.
• [13]Yiping Kang,Johann Hauswald,Cao Gao et al.Neurosurgeon:Collaborative intelli-gence between the cloud and mobile edge[J].ACM Computer Architecture News,2017,45(1):615-629.
• [14]Amir Erfan Eshratifar,Amirhossein Esmaili and Massoud Pedram.Towards collaborative intelli-gence friendly architectures for deep learning[C]//Proceedings of the International Symposium on Quality Electronic Design.2019:14-19.
• [15]Jingyue Zhou,Yihuai Wang,Kaoru Ota et al.AAIo T:Accelerating artificial intelligence in Io T systems[J].IEEE Wireless Communications Letters,2019,8(3):825-828.
• [16]Daniel Zhang,Yue Ma,Chao Zheng et al.Cooperative-competitive task allocation in edge com-puting for delay-sensitive social sensing[C]//Proceedings of the ACM/IEEE Symposium on Edge Computing.2018:243-259.
• [17]Debojyoti Hazra,Asmita Roy,Sadip Midya et al.Distributed task scheduling in cloud platform:A survey[M]//Smart computing and informatics.Springer,2018:183-191.
• [18]任杰,高岭,于佳龙等.面向边缘设备的高能效深度学习任务调度策略[J].计算机学报,2020,043(003):440-452.
• [19]Qi Cao,Zhi-Bo Wei and Wen-Mao Gong.An optimized algorithm for task scheduling based on activity based costing in cloud computing[C]//Proceedings of the International Conference on Bioinformatics and Biomedical Engineering.i CBBE,2009:1-3.
• [20]S Selvarani and G Sudha Sadhasivam.Improved cost-based algorithm for task scheduling in cloud computing[C]//Proceedings of the IEEE International Conference on Computational Intelligence and Computing Research.IEEE,2010:1-5.
• [21]Shlomo Karhi and Dvir Shabtay.Online scheduling of two job types on a set of multipurpose machines[J].International Journal of Production Economics,2014,150:155-162.
• [22]Li Guan,Weidong Li and Man Xiao.Online algorithms for the mixed ring loading problem with two nodes[J].Optimization Letters,2020:1-11.
• [23]Weidong Li,Jianping Li and Tongquan Zhang.Two approximation schemes for scheduling on parallel machines under a grade of service provision.Asia-Pacific Journal of Operational Research,2012,29(05):1250029.
• [24]Khurram Soomro,Amir Roshan Zamir and Mubarak Shah.UCF101:A dataset of 101 human actions classes from videos in the wild[J].Computer Science,2012.
• [25]Zongqing Lu,Swati Rallapalli,Kevin S Chan et al.Augur:Modeling the resource requirements of Conv Nets on mobile devices[J].IEEE Transactions on Mobile Computing,2021,20(2):352-365.