Laser & Optoelectronics Progress, Volume. 61, Issue 4, 0437002(2024)

Micro-Video Event Detection Based on Deep Dynamic Semantic Correlation

Peiguang Jing*, Xiaoyi Song, and Yuting Su
Author Affiliations
  • School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
  • show less
    Figures & Tables(9)
    Event detection method based on deep dynamic semantic correlation
    FIE module
    Dynamic correlation unit
    Convergence performance of the model
    Influence of parameters on model performance. (a) Influence of the number of hidden attributes; (b) influence of the number of important frames
    Influence of parameters on model performance. (a) Influence of α and β; (b) influence of γ
    • Table 1. Influence of different modules on the experimental results

      View table

      Table 1. Influence of different modules on the experimental results

      ModuleRPmAP
      FIE0.7230.7080.754
      FIE+ISAE0.7290.7150.768
      FIE+ISAE+DCU0.7310.7190.772
      FIE+ISAE+HAAU0.7420.7360.783
      FIE+HAAU+DCU0.7630.7540.794
      ISAE+HAAU+DCU0.7120.6970.759
      Ours0.7740.7690.817
    • Table 2. Performance comparison of different methods on the Flickr dataset

      View table

      Table 2. Performance comparison of different methods on the Flickr dataset

      MethodRPmAPTraining time /s
      C3D0.5520.6300.66316560.490
      ResNet3D0.6320.6720.69436720.080
      SRRS0.6170.6250.65691440.120
      DTSL0.6220.6720.67983160.370
      NI-SVM0.6910.7270.7493048.720
      RegNet0.7530.7490.7854505.810
      KPGNN0.7050.7310.7562856.950
      TSN0.7270.7310.7622568.560
      Gated-VIGAT0.7330.7460.7782603.050
      EventGraph0.7360.7250.7662516.970
      EITEST0,7530.7500.7842478.230
      PP-GCN0.7300.7470.7722756.280
      TSM0.7320.7260.7572586.630
      EfficientNet0.7450.7520.7812441.970
      X3D0.7610.7570.7962391.710
      Ours0.7740.7690.8172370.560
    • Table 3. Comparison of mAP of different methods on UCF101 and HMDB51 datasets

      View table

      Table 3. Comparison of mAP of different methods on UCF101 and HMDB51 datasets

      MethodUCF-101HMDB51
      C3D0.7850.533
      ResNet3D0.8140.572
      SRRS0.7630.517
      DTSL0.7910.556
      NI-SVM0.8270.573
      RegNet0.8570.595
      KPGNN0.8320.578
      TSN0.8390.581
      Gated-VIGAT0.8480.586
      EventGraph0.8420.581
      EITEST0.8460.575
      PP-GCN0.8500.599
      TSM0.8380.577
      EfficientNet0.8560.595
      X3D0.8670.608
      Ours0.8850.622
    Tools

    Get Citation

    Copy Citation Text

    Peiguang Jing, Xiaoyi Song, Yuting Su. Micro-Video Event Detection Based on Deep Dynamic Semantic Correlation[J]. Laser & Optoelectronics Progress, 2024, 61(4): 0437002

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Digital Image Processing

    Received: Mar. 30, 2023

    Accepted: Jun. 1, 2023

    Published Online: Feb. 26, 2024

    The Author Email: Peiguang Jing (pgjing@tju.edu.cn)

    DOI:10.3788/LOP230994

    CSTR:32186.14.LOP230994

    Topics