Latency and throughput characterization of convolutional neural networks for mobile computer vision

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

Standard

Latency and throughput characterization of convolutional neural networks for mobile computer vision. / Hanhirova, Jussi; Kämäräinen, Teemu; Seppälä, Sipi; Siekkinen, Matti; Hirvisalo, Vesa; Ylä-Jääski, Antti.

Proceedings of the 9th ACM Multimedia Systems Conference, MMSys 2018. ACM, 2018. p. 204-215.

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

Harvard

Hanhirova, J, Kämäräinen, T, Seppälä, S, Siekkinen, M, Hirvisalo, V & Ylä-Jääski, A 2018, Latency and throughput characterization of convolutional neural networks for mobile computer vision. in Proceedings of the 9th ACM Multimedia Systems Conference, MMSys 2018. ACM, pp. 204-215, ACM Multimedia Systems Conference, Amsterdam, Netherlands, 12/06/2018. https://doi.org/10.1145/3204949.3204975

APA

Vancouver

Author

Bibtex - Download

@inproceedings{b725cc6a0b1940518b2e71a203cd58bd,
title = "Latency and throughput characterization of convolutional neural networks for mobile computer vision",
abstract = "We study performance characteristics of convolutional neural networks (CNN) for mobile computer vision systems. CNNs have proven to be a powerful and efficient approach to implement such systems. However, the system performance depends largely on the utilization of hardware accelerators, which are able to speed up the execution of the underlying mathematical operations tremendously through massive parallelism. Our contribution is performance characterization of multiple CNN-based models for object recognition and detection with several different hardware platforms and software frameworks, using both local (on-device) and remote (network-side server) computation. The measurements are conducted using real workloads and real processing platforms. On the platform side, we concentrate especially on TensorFlow and TensorRT. Our measurements include embedded processors found on mobile devices and high-performance processors that can be used on the network side of mobile systems. We show that there exists significant latency-throughput trade-offs but the behavior is very complex. We demonstrate and discuss several factors that affect the performance and yield this complex behavior.",
author = "Jussi Hanhirova and Teemu K{\"a}m{\"a}r{\"a}inen and Sipi Sepp{\"a}l{\"a} and Matti Siekkinen and Vesa Hirvisalo and Antti Yl{\"a}-J{\"a}{\"a}ski",
year = "2018",
month = "6",
day = "12",
doi = "10.1145/3204949.3204975",
language = "English",
pages = "204--215",
booktitle = "Proceedings of the 9th ACM Multimedia Systems Conference, MMSys 2018",
publisher = "ACM",

}

RIS - Download

TY - GEN

T1 - Latency and throughput characterization of convolutional neural networks for mobile computer vision

AU - Hanhirova, Jussi

AU - Kämäräinen, Teemu

AU - Seppälä, Sipi

AU - Siekkinen, Matti

AU - Hirvisalo, Vesa

AU - Ylä-Jääski, Antti

PY - 2018/6/12

Y1 - 2018/6/12

N2 - We study performance characteristics of convolutional neural networks (CNN) for mobile computer vision systems. CNNs have proven to be a powerful and efficient approach to implement such systems. However, the system performance depends largely on the utilization of hardware accelerators, which are able to speed up the execution of the underlying mathematical operations tremendously through massive parallelism. Our contribution is performance characterization of multiple CNN-based models for object recognition and detection with several different hardware platforms and software frameworks, using both local (on-device) and remote (network-side server) computation. The measurements are conducted using real workloads and real processing platforms. On the platform side, we concentrate especially on TensorFlow and TensorRT. Our measurements include embedded processors found on mobile devices and high-performance processors that can be used on the network side of mobile systems. We show that there exists significant latency-throughput trade-offs but the behavior is very complex. We demonstrate and discuss several factors that affect the performance and yield this complex behavior.

AB - We study performance characteristics of convolutional neural networks (CNN) for mobile computer vision systems. CNNs have proven to be a powerful and efficient approach to implement such systems. However, the system performance depends largely on the utilization of hardware accelerators, which are able to speed up the execution of the underlying mathematical operations tremendously through massive parallelism. Our contribution is performance characterization of multiple CNN-based models for object recognition and detection with several different hardware platforms and software frameworks, using both local (on-device) and remote (network-side server) computation. The measurements are conducted using real workloads and real processing platforms. On the platform side, we concentrate especially on TensorFlow and TensorRT. Our measurements include embedded processors found on mobile devices and high-performance processors that can be used on the network side of mobile systems. We show that there exists significant latency-throughput trade-offs but the behavior is very complex. We demonstrate and discuss several factors that affect the performance and yield this complex behavior.

UR - http://www.scopus.com/inward/record.url?scp=85050684241&partnerID=8YFLogxK

U2 - 10.1145/3204949.3204975

DO - 10.1145/3204949.3204975

M3 - Conference contribution

SP - 204

EP - 215

BT - Proceedings of the 9th ACM Multimedia Systems Conference, MMSys 2018

PB - ACM

ER -

ID: 30194294