Optics and Precision Engineering, Volume. 30, Issue 3, 340(2022)

Multi-stage boundary reference network for action segmentation

Lin MAO, Zhe CAO*, Dawei YANG, and Rubo ZHANG
Author Affiliations
  • School of Electromechanical Engineering, Dalian Minzu University, Dalian116600, China
  • show less
    Figures & Tables(13)
    Weight calculation method
    Weight-adjustment unit structure
    WLBP structure
    Single stage boundary reference structure
    MBNet structure
    Test of convolution layer number
    Comparison of weight-adjustment boundary coefficient before and after adjustment
    Comparison of segmentation results
    • Table 1. Introduction of dataset

      View table
      View in Article

      Table 1. Introduction of dataset

      数据集视频个数动作类数平均时长/min

      GTEA

      50Salads

      Breakfast

      28

      50

      1 712

      11

      17

      48

      1.5

      7.5

      2.7

    • Table 2. Comparison of LBP and WLBP

      View table
      View in Article

      Table 2. Comparison of LBP and WLBP

      MethodF1EditAcc
      102550

      Baseline

      Multi-LBP

      Multi-WLBP

      82.3

      83.0

      83.0

      81.3

      81.9

      82.4

      74.0

      76.3

      76.3

      74.3

      75.9

      76.5

      84.4

      84.7

      84.6

    • Table 3. Segmentation result comparisons on GTEA dataset

      View table
      View in Article

      Table 3. Segmentation result comparisons on GTEA dataset

      GTEAF1EditAcc
      102550
      Dilated TCN458.852.242.2-58.3
      ED-TCN472.269.356.0-64.0
      TDRN1379.274.462.774.170.1
      MS-TCN887.585.474.681.479.2
      MS-TCN++988.885.776.083.580.1
      BCN1088.587.177.384.479.8
      Ours90.689.278.786.480.2
    • Table 4. Segmentation result comparisons on 50Salads dataset

      View table
      View in Article

      Table 4. Segmentation result comparisons on 50Salads dataset

      50SaladsF1EditAcc
      102550
      Dilated TCN452.247.637.443.159.3
      ST-CNN1455.949.637.145.959.4
      ED-TCN468.063.952.659.864.7
      TDRN1372.968.557.266.068.1
      MS-TCN876.374.064.567.980.7
      MS-TCN++980.778.570.174.383.7
      BCN1082.381.374.074.384.4
      Ours83.082.476.376.584.6
    • Table 5. Segmentation result comparisons on the Breakfast dataset

      View table
      View in Article

      Table 5. Segmentation result comparisons on the Breakfast dataset

      BreakfastF1EditAcc
      102550
      ED-TCN4----43.3
      HTK15----56.3
      GRU16----60.6
      MS-TCN852.648.137.961.766.3
      MS-TCN++964.158.645.965.667.6
      BCN1068.765.555.066.270.4
      Ours69.566.856.667.171.2
    Tools

    Get Citation

    Copy Citation Text

    Lin MAO, Zhe CAO, Dawei YANG, Rubo ZHANG. Multi-stage boundary reference network for action segmentation[J]. Optics and Precision Engineering, 2022, 30(3): 340

    Download Citation

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

    Category: Information Sciences

    Received: Apr. 20, 2021

    Accepted: --

    Published Online: Mar. 4, 2022

    The Author Email: Zhe CAO (cao_zhe@foxmail.com)

    DOI:10.37188/OPE.20223003.0340

    Topics