PokerNet: Expanding Features Cheaply via Depthwise Convolutions

doi:10.1007/s11633-021-1288-x

CASIA OpenIR > 学术期刊 > Machine Intelligence Research

	PokerNet: Expanding Features Cheaply via Depthwise Convolutions
	Wei Tang1,2 ; Yan Huang1,2 ; Liang Wang1,2
发表期刊	International Journal of Automation and Computing
ISSN	1476-8186
	2021
卷号	18 期号:3 页码:432-442
摘要	Pointwise convolution is usually utilized to expand or squeeze features in modern lightweight deep models. However, it takes up most of the overall computational cost (usually more than 90%). This paper proposes a novel Poker module to expand features by taking advantage of cheap depthwise convolution. As a result, the Poker module can greatly reduce the computational cost, and meanwhile generate a large number of effective features to guarantee the performance. The proposed module is standardized and can be employed wherever the feature expansion is needed. By varying the stride and the number of channels, different kinds of bottlenecks are designed to plug the proposed Poker module into the network. Thus, a lightweight model can be easily assembled. Experiments conducted on benchmarks reveal the effectiveness of our proposed Poker module. And our PokerNet models can reduce the computational cost by 7.1%−15.6%. PokerNet models achieve comparable or even higher recognition accuracy than previous state-of-the-art (SOTA) models on the ImageNet ILSVRC2012 classification dataset. Code is available at https://github.com/diaomin/pokernet.
关键词	Deep learning depthwise convolution lightweight deep model model compression model acceleration
DOI	10.1007/s11633-021-1288-x
引用统计	被引频次：2[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://ir.ia.ac.cn/handle/173211/44292
专题	学术期刊_Machine Intelligence Research 智能感知与计算研究中心
作者单位	1.National Laboratory of Pattern Recognition (NLPR), Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China 2.Center for Research on Intelligent Perception and Computing (CRIPAC), Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China
第一作者单位	模式识别国家重点实验室; 中国科学院自动化研究所
推荐引用方式 GB/T 7714	Wei Tang,Yan Huang,Liang Wang. PokerNet: Expanding Features Cheaply via Depthwise Convolutions[J]. International Journal of Automation and Computing,2021,18(3):432-442.
APA	Wei Tang,Yan Huang,&Liang Wang.(2021).PokerNet: Expanding Features Cheaply via Depthwise Convolutions.International Journal of Automation and Computing,18(3),432-442.
MLA	Wei Tang,et al."PokerNet: Expanding Features Cheaply via Depthwise Convolutions".International Journal of Automation and Computing 18.3(2021):432-442.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
IJAC-2020-12-337.pdf（1163KB）	期刊论文	出版稿	开放获取	CC BY-NC-SA	浏览