eduzhai > Applied Sciences > Engineering >

Neural Compression and Filtering for Edge-assisted Real-time Object Detection in Challenged Networks

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: The edge computing paradigm places compute-capable devices - edge servers -at the network edge to assist mobile devices in executing data analysis tasks.Intuitively, offloading compute-intense tasks to edge servers can reduce theirexecution time. However, poor conditions of the wireless channel connecting themobile devices to the edge servers may degrade the overall capture-to-outputdelay achieved by edge offloading. Herein, we focus on edge computingsupporting remote object detection by means of Deep Neural Networks (DNNs), anddevelop a framework to reduce the amount of data transmitted over the wirelesslink. The core idea we propose builds on recent approaches splitting DNNs intosections - namely head and tail models - executed by the mobile device and edgeserver, respectively. The wireless link, then, is used to transport the outputof the last layer of the head model to the edge server, instead of the DNNinput. Most prior work focuses on classification tasks and leaves the DNNstructure unaltered. Herein, our focus is on DNNs for three different objectdetection tasks, which present a much more convoluted structure, and modify thearchitecture of the network to: (i) achieve in-network compression byintroducing a bottleneck layer in the early layers on the head model, and (ii)prefilter pictures that do not contain objects of interest using aconvolutional neural network. Results show that the proposed techniquerepresents an effective intermediate option between local and edge computing ina parameter region where these extreme point solutions fail to providesatisfactory performance. The code and trained models are available atthis https URL .

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...