Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 8, 1107(2023)

Lightweight human pose estimation based on adaptive feature sensing

Ning WU1, Peng WANG2、*, Xiao-yan LI3, Zhi-gang LÜ3, and Meng-yu SUN4
Author Affiliations
  • 1College of Ordnance Science and Technology, Xi'an Technological University, Xi'an 710021, China
  • 2Development Planning Office, Xi'an Technological University, Xi'an 710021, China
  • 3College of Electronics and Information Engineering, Xi'an Technological University, Xi'an 710021, China
  • 4College of Optoelectronic Engineering, Xi'an Technological University, Xi'an 710021, China
  • show less
    Figures & Tables(11)
    Multi-person pose estimation network RMPE
    RMPE-tiny network structure
    Conventional convolution and Ghost convolution
    Lightweight adaptive feature perception module
    Schematic diagram of key points of the human body
    Comparison of network training losses
    Heatmap visualization analysis
    Visualization results
    • Table 1. Performance comparison of different human posture estimation methods

      View table
      View in Article

      Table 1. Performance comparison of different human posture estimation methods

      MethodsParams/MGFLOPsModel size/MBAPAP50AP75APMAPL
      CPN27.06.268.6
      SimpleBaseline-5034.08.9129.070.488.678.367.177.2
      SimpleBaseline-10153.012.4202.071.489.379.368.178.1
      HRNet28.57.1116.273.489.580.770.280.1
      HigherHRNet28.647.9109.866.487.572.861.274.2
      RMPE40.531.4162.672.389.279.169.078.6
      RMPE-tiny17.8↓22.721.2↓10.272.1↓90.572.8↑0.591.4↑2.280.1↑1.072.1↑3.179.5↑0.9
    • Table 2. Comparative analysis of ablation experiments

      View table
      View in Article

      Table 2. Comparative analysis of ablation experiments

      MethodsParams/MGFLOPsModel size/MBAPAP50AP75APMAPL
      RMPE40.531.49162.672.389.279.169.078.6
      RMPE+Ghost29.021.25116.771.791.479.972.376.5
      RMPE+Ghost+ECA17.821.2472.171.591.179.871.976.2
      RMPE+Ghost+Sa-ECA17.821.2472.172.691.580.272.079.1
      RMPE+Ghost+Sa-ECA+Huber Loss17.821.2472.172.891.480.172.179.5
    • Table 3. Model performance comparison

      View table
      View in Article

      Table 3. Model performance comparison

      模型AP检测速率/FPS
      RMPE72.37.6
      RMPE-tiny72.815.3
    Tools

    Get Citation

    Copy Citation Text

    Ning WU, Peng WANG, Xiao-yan LI, Zhi-gang LÜ, Meng-yu SUN. Lightweight human pose estimation based on adaptive feature sensing[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(8): 1107

    Download Citation

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

    Category: Research Articles

    Received: Oct. 23, 2022

    Accepted: --

    Published Online: Oct. 9, 2023

    The Author Email: Peng WANG (wp_xatu@163.com)

    DOI:10.37188/CJLCD.2022-0351

    Topics