Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications

Zheng QinZhaoning ZhangShiqing ZhangHao YuYuxing Peng

   Papers with code   Abstract  PDF

Compact neural networks are inclined to exploit "sparsely-connected" convolutions such as depthwise convolution and group convolution for employment in mobile applications. Compared with standard "fully-connected" convolutions, these convolutions are more computationally economical... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
456-MENet-24x1
(g=3)
74.7%
--
2
348-MENet-12x1
(g=3)
71.8%
--
3
352-MENet-12x1
(g=8)
68.4%
--
4
256-MENet-12x1
(g=4)
67.3%
--
5
228-MENet-12x1
(g=3)
65.9%
--
6
128-MENet-8x1
(g=4)
57.6%
--
7
160-MENet-8x1
(g=8)
56.2%
--
8
108-MENet-8x1
(g=3)
56.1%
--