PZnet: Efficient 3D ConvNet Inference on Manycore CPUs

Sergiy Popovych, Davit Buniatyan, Aleksandar Zlateski, Kai Li, Hyunjune Sebastian Seung

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Convolutional nets have been shown to achieve state-of-the-art accuracy in many biomedical image analysis tasks. Many tasks within biomedical analysis domain involve analyzing volumetric (3D) data acquired by CT, MRI and Microscopy acquisition methods. To deploy convolutional nets in practical working systems, it is important to solve the efficient inference problem. Namely, one should be able to apply an already-trained convolutional network to many large images using limited computational resources. In this paper we present PZnet, a CPU-only engine that can be used to perform inference for a variety of 3D convolutional net architectures. PZNet outperforms MKL-based CPU implementations of PyTorch and Tensorflow by more than 3.5x for the popular U-net architecture. Moreover, for 3D convolutions with low featuremap numbers, cloud CPU inference with PZnet outperforms cloud GPU inference in terms of cost efficiency.

Original languageEnglish (US)
Title of host publicationAdvances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC
EditorsSupriya Kapoor, Kohei Arai
PublisherSpringer Verlag
Pages369-383
Number of pages15
ISBN (Print)9783030177942
DOIs
StatePublished - Jan 1 2020
EventComputer Vision Conference, CVC 2019 - Las Vegas, United States
Duration: Apr 25 2019Apr 26 2019

Publication series

NameAdvances in Intelligent Systems and Computing
Volume943

Conference

ConferenceComputer Vision Conference, CVC 2019
CountryUnited States
CityLas Vegas
Period4/25/194/26/19

Fingerprint

Program processors
Convolution
Magnetic resonance imaging
Image analysis
Microscopic examination
Engines
Costs
Graphics processing unit

All Science Journal Classification (ASJC) codes

  • Control and Systems Engineering
  • Computer Science(all)

Cite this

Popovych, S., Buniatyan, D., Zlateski, A., Li, K., & Seung, H. S. (2020). PZnet: Efficient 3D ConvNet Inference on Manycore CPUs. In S. Kapoor, & K. Arai (Eds.), Advances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC (pp. 369-383). (Advances in Intelligent Systems and Computing; Vol. 943). Springer Verlag. https://doi.org/10.1007/978-3-030-17795-9_27
Popovych, Sergiy ; Buniatyan, Davit ; Zlateski, Aleksandar ; Li, Kai ; Seung, Hyunjune Sebastian. / PZnet : Efficient 3D ConvNet Inference on Manycore CPUs. Advances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC. editor / Supriya Kapoor ; Kohei Arai. Springer Verlag, 2020. pp. 369-383 (Advances in Intelligent Systems and Computing).
@inproceedings{a35776c36ad34df7b80c0cdb153fad6c,
title = "PZnet: Efficient 3D ConvNet Inference on Manycore CPUs",
abstract = "Convolutional nets have been shown to achieve state-of-the-art accuracy in many biomedical image analysis tasks. Many tasks within biomedical analysis domain involve analyzing volumetric (3D) data acquired by CT, MRI and Microscopy acquisition methods. To deploy convolutional nets in practical working systems, it is important to solve the efficient inference problem. Namely, one should be able to apply an already-trained convolutional network to many large images using limited computational resources. In this paper we present PZnet, a CPU-only engine that can be used to perform inference for a variety of 3D convolutional net architectures. PZNet outperforms MKL-based CPU implementations of PyTorch and Tensorflow by more than 3.5x for the popular U-net architecture. Moreover, for 3D convolutions with low featuremap numbers, cloud CPU inference with PZnet outperforms cloud GPU inference in terms of cost efficiency.",
author = "Sergiy Popovych and Davit Buniatyan and Aleksandar Zlateski and Kai Li and Seung, {Hyunjune Sebastian}",
year = "2020",
month = "1",
day = "1",
doi = "https://doi.org/10.1007/978-3-030-17795-9_27",
language = "English (US)",
isbn = "9783030177942",
series = "Advances in Intelligent Systems and Computing",
publisher = "Springer Verlag",
pages = "369--383",
editor = "Supriya Kapoor and Kohei Arai",
booktitle = "Advances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC",
address = "Germany",

}

Popovych, S, Buniatyan, D, Zlateski, A, Li, K & Seung, HS 2020, PZnet: Efficient 3D ConvNet Inference on Manycore CPUs. in S Kapoor & K Arai (eds), Advances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC. Advances in Intelligent Systems and Computing, vol. 943, Springer Verlag, pp. 369-383, Computer Vision Conference, CVC 2019, Las Vegas, United States, 4/25/19. https://doi.org/10.1007/978-3-030-17795-9_27

PZnet : Efficient 3D ConvNet Inference on Manycore CPUs. / Popovych, Sergiy; Buniatyan, Davit; Zlateski, Aleksandar; Li, Kai; Seung, Hyunjune Sebastian.

Advances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC. ed. / Supriya Kapoor; Kohei Arai. Springer Verlag, 2020. p. 369-383 (Advances in Intelligent Systems and Computing; Vol. 943).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - PZnet

T2 - Efficient 3D ConvNet Inference on Manycore CPUs

AU - Popovych, Sergiy

AU - Buniatyan, Davit

AU - Zlateski, Aleksandar

AU - Li, Kai

AU - Seung, Hyunjune Sebastian

PY - 2020/1/1

Y1 - 2020/1/1

N2 - Convolutional nets have been shown to achieve state-of-the-art accuracy in many biomedical image analysis tasks. Many tasks within biomedical analysis domain involve analyzing volumetric (3D) data acquired by CT, MRI and Microscopy acquisition methods. To deploy convolutional nets in practical working systems, it is important to solve the efficient inference problem. Namely, one should be able to apply an already-trained convolutional network to many large images using limited computational resources. In this paper we present PZnet, a CPU-only engine that can be used to perform inference for a variety of 3D convolutional net architectures. PZNet outperforms MKL-based CPU implementations of PyTorch and Tensorflow by more than 3.5x for the popular U-net architecture. Moreover, for 3D convolutions with low featuremap numbers, cloud CPU inference with PZnet outperforms cloud GPU inference in terms of cost efficiency.

AB - Convolutional nets have been shown to achieve state-of-the-art accuracy in many biomedical image analysis tasks. Many tasks within biomedical analysis domain involve analyzing volumetric (3D) data acquired by CT, MRI and Microscopy acquisition methods. To deploy convolutional nets in practical working systems, it is important to solve the efficient inference problem. Namely, one should be able to apply an already-trained convolutional network to many large images using limited computational resources. In this paper we present PZnet, a CPU-only engine that can be used to perform inference for a variety of 3D convolutional net architectures. PZNet outperforms MKL-based CPU implementations of PyTorch and Tensorflow by more than 3.5x for the popular U-net architecture. Moreover, for 3D convolutions with low featuremap numbers, cloud CPU inference with PZnet outperforms cloud GPU inference in terms of cost efficiency.

UR - http://www.scopus.com/inward/record.url?scp=85065502139&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85065502139&partnerID=8YFLogxK

U2 - https://doi.org/10.1007/978-3-030-17795-9_27

DO - https://doi.org/10.1007/978-3-030-17795-9_27

M3 - Conference contribution

SN - 9783030177942

T3 - Advances in Intelligent Systems and Computing

SP - 369

EP - 383

BT - Advances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC

A2 - Kapoor, Supriya

A2 - Arai, Kohei

PB - Springer Verlag

ER -

Popovych S, Buniatyan D, Zlateski A, Li K, Seung HS. PZnet: Efficient 3D ConvNet Inference on Manycore CPUs. In Kapoor S, Arai K, editors, Advances in Computer Vision - Proceedings of the 2019 Computer Vision Conference CVC. Springer Verlag. 2020. p. 369-383. (Advances in Intelligent Systems and Computing). https://doi.org/10.1007/978-3-030-17795-9_27