Adaptive Spectral Graph Convolutional Networks for Skeleton-Based Action Recognition
share
Traditional deep methods for skeleton-based action recognition usually structure the skeleton as a coordinates sequence or a pseudo-image to feed to RNNs or CNNs, which cannot explicitly exploit the natural connectivity among the joints. Recently, graph convolutional networks (GCNs), which generalize CNNs to more generic non-Euclidean structures, obtains remarkable performance for skeleton-based action recognition. In this work, we propose a novel adaptive spectral graph convolutional network which address the three major issues exposed in former GCN-based methods applied to action recognition: 1) The sampling function of graph convolutional operation applied in former methods is usually heuristically designed. We avoid the tricked definition of sampling function by transforming the GCN to frequency domain based on spectral graph theory. 2) The topology of the graph is set by hand and fixed over all layers, which may be not optimal for the action recognition task and the hierarchical CNN structures. In our model, the topology of the graph in each layer can be either uniformly or individually learned by BP algorithm, which can bring more flexibility and generality. 3) The first-order information (the coordinate of joints) is mainly used in former GCNs, while the second-order information (the length and direction of bones) is less exploited. A two-stream framework is proposed in this work to model both of the joints and bones information simultaneously. Extensive experiments on two large-scale datasets, NTU-RGB+D and Kinetics, demonstrate the performance of our model exceeds the state-of-the-art by a significant margin.
READ FULL TEXT
