• Optical Instruments
  • Vol. 44, Issue 4, 39 (2022)
Xiaofei QIN1, Ying ZHAO1, Yijie ZHANG1, Ruijie DU1..., Hanwen QIAN1, Meng CHEN2, Wenqi ZHANG2 and Xuedian ZHANG1|Show fewer author(s)
Author Affiliations
  • 1School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • 2Institute of Aerospace System Engineering of Shanghai, Shanghai 201109, China
  • show less
    DOI: 10.3969/j.issn.1005-5630.2022.004.006 Cite this Article
    Xiaofei QIN, Ying ZHAO, Yijie ZHANG, Ruijie DU, Hanwen QIAN, Meng CHEN, Wenqi ZHANG, Xuedian ZHANG. Multiscale hypergraph convolutional network for skeleton-based action recognition[J]. Optical Instruments, 2022, 44(4): 39 Copy Citation Text show less
    Action recognition process
    Fig. 1. Action recognition process
    Structure of multiscale hypergraph convolutional network
    Fig. 2. Structure of multiscale hypergraph convolutional network
    Structure of adaptive graph convolution block
    Fig. 3. Structure of adaptive graph convolution block
    Allocation strategy for hyperedge merging
    Fig. 4. Allocation strategy for hyperedge merging
    Structure of multiscale temporal graph convolution block
    Fig. 5. Structure of multiscale temporal graph convolution block
    Learning curve of multiscale hypergraph convolutional network on NTU-RGB+D dataset
    Fig. 6. Learning curve of multiscale hypergraph convolutional network on NTU-RGB+D dataset
    方法CV精度/%
    Baseline95.3
    Baseline+ ${\varepsilon }_{10}$96.0
    Baseline+ ${\varepsilon }_{10}$+ ${\varepsilon }_{5}$97.1
    Table 1. Ablation study of HCB and HMB
    结构数据CV精度/%
    三流骨架、关节、动态97.1
    w/o骨架96.0
    两流w/o关节95.8
    w/o动态95.2
    骨架93.7
    单流关节93.5
    动态92.1
    Table 2. Comparison of results obtained via different skeleton input data
    空洞率CV精度/%
    195.1
    295.6
    396.1
    495.9
    1, 2, 3, 497.1
    Table 3. The performance of models with different dilation factors
    方法CS精度/%CV精度/%
    Lie-Group[2]50.152.8
    TCN[23]74.383.1
    ST-GCN[9]86.894.2
    AS-GCN[14]86.894.2
    2s-AGCN[13]88.595.1
    SGN[17]89.094.5
    AGC-LSTM[7]89.295.0
    DGNN[24]89.996.1
    Res-GCN[11]90.096.0
    SGCN[16]90.196.2
    MHCN91.297.1
    Table 4. Comparison with state-of-the-art methods on the NTU-RGB+D dataset
    方法TOP1精度/%TOP5精度/%
    TCN[23]20.340.0
    ST-GCN[9]30.752.8
    AS-GCN[14]34.856.5
    DGNN[24]36.959.6
    2s-AGCN[13]36.158.7
    Hyper-GCN[18]37.160.0
    SGCN[16]37.160.1
    MHCN38.161.8
    Table 5. Comparison with state-of-the-art methods on the Kinetics dataset
    Xiaofei QIN, Ying ZHAO, Yijie ZHANG, Ruijie DU, Hanwen QIAN, Meng CHEN, Wenqi ZHANG, Xuedian ZHANG. Multiscale hypergraph convolutional network for skeleton-based action recognition[J]. Optical Instruments, 2022, 44(4): 39
    Download Citation