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

Jussi Hanhirova, Teemu Kämäräinen, Sipi Seppälä, Matti Siekkinen, Vesa Hirvisalo, Antti Ylä-Jääski

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

59 Citations (Scopus)


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.

Original languageEnglish
Title of host publicationProceedings of the 9th ACM Multimedia Systems Conference, MMSys 2018
Number of pages12
ISBN (Electronic)9781450351928
Publication statusPublished - 12 Jun 2018
MoE publication typeA4 Conference publication
EventACM Multimedia Systems Conference - Amsterdam, Netherlands
Duration: 12 Jun 201815 Jun 2018
Conference number: 9


ConferenceACM Multimedia Systems Conference
Abbreviated titleMMSys


Dive into the research topics of 'Latency and throughput characterization of convolutional neural networks for mobile computer vision'. Together they form a unique fingerprint.

Cite this