Neural Compression and Filtering for Edge-assisted Real-time Object Detection in Challenged Networks (2020-07-31T00:00:00.000000Z)

TL;DR

This work focuses on DNNs for three different object detection tasks, and modify the architecture of the network to achieve in-network compression by introducing a bottleneck layer in the early layers on the head model, and prefilter pictures that do not contain objects of interest using a convolutional neural network.

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 their execution time. However, poor conditions of the wireless channel connecting the mobile devices to the edge servers may degrade the overall capture-to-output delay achieved by edge offloading. Herein, we focus on edge computing supporting remote object detection by means of Deep Neural Networks (DNNs), and develop a framework to reduce the amount of data transmitted over the wireless link. The core idea we propose builds on recent approaches splitting DNNs into sections - namely head and tail models - executed by the mobile device and edge server, respectively. The wireless link, then, is used to transport the output of the last layer of the head model to the edge server, instead of the DNN input. Most prior work focuses on classification tasks and leaves the DNN structure unaltered. Herein, our focus is on DNNs for three different object detection tasks, which present a much more convoluted structure, and modify the architecture of the network to: (i) achieve in-network compression by introducing a bottleneck layer in the early layers on the head model, and (ii) prefilter pictures that do not contain objects of interest using a convolutional neural network. Results show that the proposed technique represents an effective intermediate option between local and edge computing in a parameter region where these extreme point solutions fail to provide satisfactory performance. The code and trained models are available at https://github.com/yoshitomo-matsubaralhnd-ghnd-object-detectors.

Authors

References39 items

Split Computing for Complex Object Detectors: Challenges and Preliminary Results

Fast and Accurate Streaming CNN Inference via Communication Compression on the Edge

PyTorch: An Imperative Style, High-Performance Deep Learning Library

BottleNet++: An End-to-End Approach for Feature Compression in Device-Edge Co-Inference Systems

Distilled Split Deep Neural Networks for Edge-Assisted Real-Time Systems

Cracking open the DNN black-box: Video Analytics with DNNs across the Camera-Cloud Boundary

Learning Lightweight Pedestrian Detector with Hierarchical Knowledge Distillation

Edge Assisted Real-time Object Detection for Mobile Augmented Reality

Source Compression with Bounded DNN Perception Loss for IoT Edge Computer Vision

Distilling Object Detectors With Fine-Grained Feature Imitation

BottleNet: A Deep Learning Architecture for Intelligent Mobile Cloud Computing Services

Auto-tuning Neural Network Quantization Framework for Collaborative Inference Between the Cloud and Edge

Balancing Video Analytics Processing and Bandwidth for Edge-Cloud Networks

Video Compression for Object Detection Algorithms

Collaborative Edge and Cloud Neural Networks for Real-Time Video Processing

Computation Offloading for Machine Learning Web Apps in the Edge Server Environment

Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference

Learning Efficient Object Detection Models with Knowledge Distillation

Mimicking Very Efficient Network for Object Detection

Distributed Deep Neural Networks Over the Cloud, the Edge and End Devices

Neurosurgeon: Collaborative Intelligence Between the Cloud and Mobile Edge

Mask R-CNN

Feature Pyramid Networks for Object Detection

Deep Residual Learning for Image Recognition

SSD: Single Shot MultiBox Detector

You Only Look Once: Unified, Real-Time Object Detection

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

Fast R-CNN

Distilling the Knowledge in a Neural Network

Adam: A Method for Stochastic Optimization

FitNets: Hints for Thin Deep Nets

Do Deep Nets Really Need to be Deep?

Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation

Poselets: Body part detectors trained using 3D human pose annotations

Extracting and composing robust features with denoising autoencoders

Fog computing and its role in the internet of things

of the neural ﬁlter, where the output of L0 in the “frozen” student model is fed to the neural ﬁlter

Conv2d(oc=64, k=1, s=1) Conv2d(oc=64, k=2, p BatchNorm2d BatchNorm2d Conv2d(oc=64, k=3, s=1, p=1) Conv2d(oc=256, k=2,

Architectures of Layer 1 (L1) in teacher and student models Teacher's L1 Student's L1