Abstract:Graph convolutional networks (GCNs) have achieved promising results in skeleton-based action recognition due to their capability in analysing irregular grids with non-Euclidean geometry. Considering the fact that skeleton based action recognition is a classification problem, the suitable graph representation in GCN-based approaches is the key for the ultimate goal of classification. Despite the practical success of GCN-based approaches for solving this problem over the past few years, learning better representation is still a challenging issue and the existing approaches fail in distinguishing similar actions. Besides, most existing GCN-based frameworks focus on modelling the spatial information and use one fixed kernel to model the temporal information. Such modelling does not pay enough attention to diversifying representations among different skeleton frames, leading to inefficiency in obtaining more discriminative temporal features for different actions and therefore, such modelling is inconvenient with the diversity of human movements. Our main concern in this work is the adaptive feature extraction of highly discriminative information for both the spatial and the temporal dimensions. To achieve that, a novel Enhanced Discriminative Graph Convolutional Network (ED-GCN) based on the attention mechanism for skeleton-based action recognition is proposed. Discriminative channel-wise features are obtained by fusing the Squeeze and Excitation (SE) module to the GCN to selectively enhance the significant features and suppress the non-significant ones. The adaptively enhanced feature map is then fused to the graph convolutional layer to improve the capability of learning better representation. For the temporal dimension inspired by temporal modelling in videos, we introduce our adaptive temporal modelling block (ATB), which is able to flexibly capture temporal structure for skeleton-based action recognition. Here, the proposed ATB is a two-stage module comprising re-calibration and motion-interaction stage, designed to learn temporal features by integrating the modelling of channel correlation and temporal evolution, respectively. Experimental results on two large-scale datasets, NTU-RGB+D and Kinetics-skeleton demonstrate the importance of discriminatively learned information and the effectiveness of the proposed ED-GCN for skeleton-based action recognition.

Enhanced discriminative graph convolutional network with adaptive temporal modelling for skeleton-based action recognition

Temporal‐enhanced graph convolution network for skeleton‐based action recognition

Adaptive Attention Memory Graph Convolutional Networks for Skeleton-Based Action Recognition

A Tri-Attention Enhanced Graph Convolutional Network for Skeleton-Based Action Recognition

Temporal segment graph convolutional networks for skeleton-based action recognition

Enhanced Spatial and Extended Temporal Graph Convolutional Network for Skeleton-Based Action Recognition

Temporal Refinement Graph Convolutional Network for Skeleton-based Action Recognition

Temporal-Aware Graph Convolution Network for Skeleton-based Action Recognition.

Temporal Enhanced Multi-Stream Graph Convolutional Nerual Networks For Skeleton-Based Action Recognition

Temporal Graph Modeling for Skeleton-based Action Recognition

An Attentional Spatial Temporal Graph Convolutional Network with Co-Occurrence Feature Learning for Action Recognition

Skeleton-based Action Recognition with Multi-stream, Temporal-Channel Enhanced Graph Convolution Network

Information Enhanced Graph Convolutional Networks for Skeleton-based Action Recognition

Temporal Attention-Augmented Graph Convolutional Network for Efficient Skeleton-Based Human Action Recognition

Channel-Wise Dense Connection Graph Convolutional Network for Skeleton-Based Action Recognition

Spatial-Temporal Adaptive Graph Convolutional Network for Skeleton-Based Action Recognition.

An improved spatial temporal graph convolutional network for robust skeleton-based action recognition

Skeleton-Based Action Recognition with Multi-Stream Adaptive Graph Convolutional Networks

Densely Connected and Multiple Temporal Graph Convolution Networks for Skeleton-based Action Recognition

Dual-domain Graph Convolutional Networks for Skeleton-Based Action Recognition

DSDC-GCN: Decoupled Static-Dynamic Co-occurrence Graph Convolutional Networks for Skeleton-Based Action Recognition