Graph convolutional network with structure pooling and joint-wise channel attention for action recognition
作者：
Highlights：
• We propose a novel structure based graph pooling (SGP) scheme to gradually expand the receptivefields of graph convolution kernels in deeper layers, which can enhance the ability of GCNs for extracting more global motion information and bring a reduction in the amount of parameters and computation cost.
• We propose a joint-wise channel attention (JCA) module to mine discriminative information among confusing actions with attention mechanism, which shows significant improvement for classifying confusion actions.
• Experimental results demonstrate that our model outperforms the existing state-of-the-art methods.
摘要
•We propose a novel structure based graph pooling (SGP) scheme to gradually expand the receptivefields of graph convolution kernels in deeper layers, which can enhance the ability of GCNs for extracting more global motion information and bring a reduction in the amount of parameters and computation cost.•We propose a joint-wise channel attention (JCA) module to mine discriminative information among confusing actions with attention mechanism, which shows significant improvement for classifying confusion actions.•Experimental results demonstrate that our model outperforms the existing state-of-the-art methods.
论文关键词：Graph convolutional network,Structure graph pooling,Joint-wise channel attention
论文评审过程：Received 1 June 2019, Revised 22 January 2020, Accepted 28 February 2020, Available online 29 February 2020, Version of Record 19 March 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2020.107321
