Abstract
Pooling is a crucial operation in computer vision, yet the unique structureof skeletons hinders the application of existing pooling strategies to skeletongraph modelling. In this paper, we propose an Improved Graph Pooling Network,referred to as IGPN. The main innovations include: Our method incorporates aregion-awareness pooling strategy based on structural partitioning. Thecorrelation matrix of the original feature is used to adaptively adjust theweight of information in different regions of the newly generated features,resulting in more flexible and effective processing. To prevent theirreversible loss of discriminative information, we propose a cross fusionmodule and an information supplement module to provide block-level andinput-level information respectively. As a plug-and-play structure, theproposed operation can be seamlessly combined with existing GCN-based models.We conducted extensive evaluations on several challenging benchmarks, and theexperimental results indicate the effectiveness of our proposed solutions. Forexample, in the cross-subject evaluation of the NTU-RGB+D 60 dataset, IGPNachieves a significant improvement in accuracy compared to the baseline whilereducing Flops by nearly 70%; a heavier version has also been introduced tofurther boost accuracy.