Asymmetric information-regularized learning for skeleton-based action recognition

被引:0
作者
Kunlun Wu
Xun Gong
机构
[1] Southwest Jiaotong University,School of Computing and Artificial Intelligence
来源
Applied Intelligence | 2023年 / 53卷
关键词
Skeleton-based action recognition; Graph convolutional networks; Deep learning; Asymmetric information-regularized learning;
D O I
暂无
中图分类号
学科分类号
摘要
Skeleton-based action recognition has recently achieved remarkable progress, which is typically formulated as a spatial-temporal graph-based classification problem. Nevertheless, most existing approaches straightforwardly model the skeleton topology via a pure encoder and lack explicit guidance to promote the representation capability. To handle the above constraint, the proposed Asymmetric Information-Regularized Graph Convolutional Network (AIR-GCN) explores an effective asymmetric paradigm based on information theory, to force the encoder to learn more representative features. Furthermore, each sample indeed has a unique spatial-temporal topology due to the dynamic action process and AIR-GCN introduces two novel operators to learn spatial-temporal representation beyond the inherent structural relations: leveraging the Topology-regularized Spatial Routing (TrSR) to encode instance-dependent relational graphs and the Topology-regularized Temporal Routing (TrTR) to capture action-specific motion patterns for reducing the ambiguity of highly similar actions. Extensive experiments are conducted on four widely used datasets: Northwestern-UCLA, NTU RGB+D 60, NTU RGB+D 120 and Kinetics Skeleton. The results demonstrate that AIR-GCN achieves notably better performance compared with the state-of-the-art methods.
引用
收藏
页码:31065 / 31076
页数:11
相关论文
共 48 条
  • [1] Bronstein MM(2017)Geometric deep learning: going beyond euclidean data IEEE Signal Process Mag 34 18-42
  • [2] Bruna J(2017)Enhanced skeleton visualization for view invariant human action recognition Pattern Recogn 68 346-362
  • [3] LeCun Y(2019)Ntu rgb+ d 120: A large-scale benchmark for 3d human activity understanding IEEE Trans Pattern Anal Mach Intell 42 2684-2701
  • [4] Szlam A(2013)Learning actionlet ensemble for 3d human action recognition IEEE Trans Pattern Anal Mach Intell 36 914-927
  • [5] Vandergheynst P(2021)Multi-scale spatial temporal graph convolutional network for skeleton-based action recognition In Proceedings of the AAAI Conference on Artificial Intelligence 35 1113-1122
  • [6] Liu M(2022)Topology-aware convolutional neural network for efficient skeleton-based action recognition In Proceedings of the AAAI Conference on Artificial Intelligence 36 2866-2874
  • [7] Liu H(2022)Enhanced discriminative graph convolutional network with adaptive temporal modelling for skeleton-based action recognition Comput Vis Image Underst 216 2575-2585
  • [8] Chen C(2023)Skeleton-based human action recognition via large-kernel attention graph convolutional network IEEE Trans Visual Comput Graphics 29 1474-1488
  • [9] Liu J(2023)Relation-mining self-attention network for skeleton-based human action recognition Pattern Recogn 139 undefined-undefined
  • [10] Shahroudy A(2022)Graph transformer network with temporal kernel attention for skeleton-based action recognition Knowl-Based Syst 240 undefined-undefined