ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices

Xiangyu ZhangXinyu ZhouMengxiao LinJian Sun

   Papers with code   Abstract  PDF

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... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
ShuffleNet x1.0
(g=8)
65.9%
--
2
ShuffleNet x1.0
(g=4)
65.8%
--
3
ShuffleNet x1.0
(g=2)
65.8%
--
4
ShuffleNet x1.0
(g=3)
65.6%
--
5
ShuffleNet x1.0
(g=1)
65.1%
--
6
ShuffleNet x0.75
(g=3)
61.8%
--
7
ShuffleNet x0.75
(g=1)
60.4%
--
8
ShuffleNet x0.5
(g=3)
55.8%
--
9
ShuffleNet x0.5
(g=1)
53.4%
--
10
ShuffleNet x0.25
(g=3)
38.3%
--
11
ShuffleNet x0.25
(g=1)
37.6%
--