We introduce an extremely computation-efficient CNN architecture named ShuffleNet, which is designed specially for mobile devices with very limited computing power (e.g., 10-150 MFLOPs). The new architecture utilizes two new operations, pointwise group convolution and channel shuffle, to greatly reduce computation cost while maintaining accuracy. Experiments on ImageNet classification and MS COCO object detection demonstrate the superior performance of ShuffleNet over other structures, e.g. lower top-1 error (absolute 7.8%) than recent MobileNet 12 on ImageNet classification task, under the computation budget of 40 MFLOPs. On an ARM-based mobile device, ShuffleNet achieves ~13× actual speedup over AlexNet while maintaining comparable accuracy.
Description
ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices - IEEE Conference Publication
%0 Conference Paper
%1 8578814
%A Zhang, X.
%A Zhou, X.
%A Lin, M.
%A Sun, J.
%B 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition
%D 2018
%K ShuffleNet order1
%P 6848-6856
%R 10.1109/CVPR.2018.00716
%T ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices
%U https://ieeexplore.ieee.org/document/8578814
%X We introduce an extremely computation-efficient CNN architecture named ShuffleNet, which is designed specially for mobile devices with very limited computing power (e.g., 10-150 MFLOPs). The new architecture utilizes two new operations, pointwise group convolution and channel shuffle, to greatly reduce computation cost while maintaining accuracy. Experiments on ImageNet classification and MS COCO object detection demonstrate the superior performance of ShuffleNet over other structures, e.g. lower top-1 error (absolute 7.8%) than recent MobileNet 12 on ImageNet classification task, under the computation budget of 40 MFLOPs. On an ARM-based mobile device, ShuffleNet achieves ~13× actual speedup over AlexNet while maintaining comparable accuracy.
@inproceedings{8578814,
abstract = {We introduce an extremely computation-efficient CNN architecture named ShuffleNet, which is designed specially for mobile devices with very limited computing power (e.g., 10-150 MFLOPs). The new architecture utilizes two new operations, pointwise group convolution and channel shuffle, to greatly reduce computation cost while maintaining accuracy. Experiments on ImageNet classification and MS COCO object detection demonstrate the superior performance of ShuffleNet over other structures, e.g. lower top-1 error (absolute 7.8%) than recent MobileNet [12] on ImageNet classification task, under the computation budget of 40 MFLOPs. On an ARM-based mobile device, ShuffleNet achieves ~13× actual speedup over AlexNet while maintaining comparable accuracy.},
added-at = {2020-05-11T23:14:45.000+0200},
author = {{Zhang}, X. and {Zhou}, X. and {Lin}, M. and {Sun}, J.},
biburl = {https://www.bibsonomy.org/bibtex/21a6e0a4774cda410da5b3bd836ce21d8/sohnki},
booktitle = {2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition},
description = {ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices - IEEE Conference Publication},
doi = {10.1109/CVPR.2018.00716},
interhash = {0cf3b2b8f5a0d5a872680ddf57228bf4},
intrahash = {1a6e0a4774cda410da5b3bd836ce21d8},
issn = {2575-7075},
keywords = {ShuffleNet order1},
month = {June},
pages = {6848-6856},
timestamp = {2020-06-02T20:04:23.000+0200},
title = {ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices},
url = {https://ieeexplore.ieee.org/document/8578814},
year = 2018
}