Leveraging transprecision computing for machine vision applications at the edge

Umar Ibrahim Minhas; Lev Mukhanov; Georgios Karakonstantis; Hans Vandierendonck; Roger Woods

Leveraging transprecision computing for machine vision applications at the edge

Umar Ibrahim Minhas, Lev Mukhanov, Georgios Karakonstantis, Hans Vandierendonck, Roger Woods

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Machine vision tasks presents challenges for re-source constrained edge devices, particularly as they executemultiple tasks with variable workloads. A robust approachthat can dynamically adapt in runtime while maintaining themaximum quality of service (QoS) within resource constraints, isneeded. The paper presents a lightweight approach that monitorsthe runtime workload constraints, leverages accuracy-throughputtrade-off and optimisation techniques to find configurations foreach task for optimal accuracy, energy and memory and managestransparent switching between configurations. For an accuracydrop of 1%, we show a1.6×higher achieved frame processingrate with further improvements possible at lower accuracy.

Original language	English
Title of host publication	2021 IEEE workshop on signal processing systems (SiPS): proceedings
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	205-210
ISBN (Electronic)	9781665401449
ISBN (Print)	9781665401456
Publication status	Published - 11 Nov 2021
Event	2021 IEEE Workshop on Signal Processing Systems (SiPS) - Coimbra, Portugal Duration: 19 Oct 2021 → 21 Oct 2021

Publication series

Name	IEEE Workshop on Signal Processing Systems (SiPS)
Publisher	IEEE
ISSN (Print)	1520-6130
ISSN (Electronic)	2374-7390

Conference

Conference	2021 IEEE Workshop on Signal Processing Systems (SiPS)
Country/Territory	Portugal
City	Coimbra
Period	19/10/2021 → 21/10/2021

Cite this

@inproceedings{f6dc8b0435dc4424958b84f0c5082c81,

title = "Leveraging transprecision computing for machine vision applications at the edge",

abstract = "Machine vision tasks presents challenges for re-source constrained edge devices, particularly as they executemultiple tasks with variable workloads. A robust approachthat can dynamically adapt in runtime while maintaining themaximum quality of service (QoS) within resource constraints, isneeded. The paper presents a lightweight approach that monitorsthe runtime workload constraints, leverages accuracy-throughputtrade-off and optimisation techniques to find configurations foreach task for optimal accuracy, energy and memory and managestransparent switching between configurations. For an accuracydrop of 1%, we show a1.6×higher achieved frame processingrate with further improvements possible at lower accuracy.",

author = "Minhas, {Umar Ibrahim} and Lev Mukhanov and Georgios Karakonstantis and Hans Vandierendonck and Roger Woods",

year = "2021",

month = nov,

day = "11",

language = "English",

isbn = "9781665401456",

series = "IEEE Workshop on Signal Processing Systems (SiPS)",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "205--210",

booktitle = "2021 IEEE workshop on signal processing systems (SiPS): proceedings",

address = "United States",

note = "2021 IEEE Workshop on Signal Processing Systems (SiPS) ; Conference date: 19-10-2021 Through 21-10-2021",

}

Minhas, UI, Mukhanov, L, Karakonstantis, G , Vandierendonck, H & Woods, R 2021, Leveraging transprecision computing for machine vision applications at the edge. in 2021 IEEE workshop on signal processing systems (SiPS): proceedings. IEEE Workshop on Signal Processing Systems (SiPS), Institute of Electrical and Electronics Engineers Inc., pp. 205-210, 2021 IEEE Workshop on Signal Processing Systems (SiPS), Coimbra, Portugal, 19/10/2021.

Leveraging transprecision computing for machine vision applications at the edge. / Minhas, Umar Ibrahim; Mukhanov, Lev; Karakonstantis, Georgios et al.
2021 IEEE workshop on signal processing systems (SiPS): proceedings. Institute of Electrical and Electronics Engineers Inc., 2021. p. 205-210 (IEEE Workshop on Signal Processing Systems (SiPS)).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Leveraging transprecision computing for machine vision applications at the edge

AU - Minhas, Umar Ibrahim

AU - Mukhanov, Lev

AU - Karakonstantis, Georgios

AU - Vandierendonck, Hans

AU - Woods, Roger

PY - 2021/11/11

Y1 - 2021/11/11

N2 - Machine vision tasks presents challenges for re-source constrained edge devices, particularly as they executemultiple tasks with variable workloads. A robust approachthat can dynamically adapt in runtime while maintaining themaximum quality of service (QoS) within resource constraints, isneeded. The paper presents a lightweight approach that monitorsthe runtime workload constraints, leverages accuracy-throughputtrade-off and optimisation techniques to find configurations foreach task for optimal accuracy, energy and memory and managestransparent switching between configurations. For an accuracydrop of 1%, we show a1.6×higher achieved frame processingrate with further improvements possible at lower accuracy.

AB - Machine vision tasks presents challenges for re-source constrained edge devices, particularly as they executemultiple tasks with variable workloads. A robust approachthat can dynamically adapt in runtime while maintaining themaximum quality of service (QoS) within resource constraints, isneeded. The paper presents a lightweight approach that monitorsthe runtime workload constraints, leverages accuracy-throughputtrade-off and optimisation techniques to find configurations foreach task for optimal accuracy, energy and memory and managestransparent switching between configurations. For an accuracydrop of 1%, we show a1.6×higher achieved frame processingrate with further improvements possible at lower accuracy.

M3 - Conference contribution

SN - 9781665401456

T3 - IEEE Workshop on Signal Processing Systems (SiPS)

SP - 205

EP - 210

BT - 2021 IEEE workshop on signal processing systems (SiPS): proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2021 IEEE Workshop on Signal Processing Systems (SiPS)

Y2 - 19 October 2021 through 21 October 2021

ER -

Leveraging transprecision computing for machine vision applications at the edge

Abstract

Publication series

Conference

Fingerprint

Cite this