Context extraction module for deep convolutional neural networks
作者：
Highlights：
• A novel approach to improve pointwise convolution by using context information.
• Context information extracted and refined at multiple scales.
• Improves network performance without significant increase in parameters/computation.
• Networks modified with our approach exhibit improved convergence characteristics.
• Also boosts performance of other performance improvement techniques.
摘要
•A novel approach to improve pointwise convolution by using context information.•Context information extracted and refined at multiple scales.•Improves network performance without significant increase in parameters/computation.•Networks modified with our approach exhibit improved convergence characteristics.•Also boosts performance of other performance improvement techniques.
论文关键词：Contextual pointwise convolution,Convolutional neural network (CNN),Image classification,Deep learning
论文评审过程：Received 29 July 2020, Revised 12 August 2021, Accepted 28 August 2021, Available online 30 August 2021, Version of Record 17 September 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.108284
