Optics and Precision Engineering, Volume. 31, Issue 12, 1859(2023)

Few-shot object detection on Thangka via multi-scale context information

Wenjin HU1,2、*, Huiyuan TANG1, Chaoyang YUE1, and Huafei SONG1
Author Affiliations
  • 1Key Laboratory of China's Ethnic Languages and Information Technology Ministry of Education, Northwest Minzu University,Lanzhou730030, China
  • 2School of Mathematics and Computer Science, Northwest Minzu University,Lanzhou730030,China
  • show less
    Figures & Tables(12)
    FSMC network model
    Contextual feature extraction module
    Ordinary convolution and dilated convolution
    Dual attention module
    Spatial attention module
    Channel attention module
    Comparison of loss function curves
    Experimental results
    • Table 1. Comparison of detection results of different algorithms in Thangka dataset

      View table
      View in Article

      Table 1. Comparison of detection results of different algorithms in Thangka dataset

      ShotMethodsAP50AP75mAP
      10LSTD1211.39.710.3
      TFA1414.29.710.8
      MPSR1915.513.613.2
      FSCE1529.618.016.7
      FSMC34.619.619.7
      20LSTD1214.511.712.3
      TFA1416.11.514.3
      MPSR1921.716.114.6
      FSCE1535.619.218.1
      FSMC41.825.724.4
    • Table 2. Detection results of different categories

      View table
      View in Article

      Table 2. Detection results of different categories

      shotmethodssumeruSeatpadmasananaturesolemnInstrumentchignoncoronet
      10FSCE11.710.017.614.231.035.8
      FSMC17.314.618.519.332.137.1
      20FSCE13.913.622.419.233.136.3
      FSMC21.418.122.525.533.536.5
    • Table 3. Comparison of detection results of COCO dataset

      View table
      View in Article

      Table 3. Comparison of detection results of COCO dataset

      ShotMethodsAP50AP75mAP
      10FSCE10.47.28.9
      FSMC12.110.611.2
      20FSCE12.512.711.5
      FSMC15.614.414.9
    • Table 4. Ablation experiment results on Thangka dataset

      View table
      View in Article

      Table 4. Ablation experiment results on Thangka dataset

      ShotMethodsAP50AP75mAP

      Parameters/

      104(M)

      Average detection time/s
      10FSCE29.618.016.71.120.050
      FSCE+CFM32.618.218.0--
      FSCE+CFM+DAM33.919.119.4--
      FSCE+CFM+DAM+RSLoss (FSMC)34.619.619.71.170.093
    Tools

    Get Citation

    Copy Citation Text

    Wenjin HU, Huiyuan TANG, Chaoyang YUE, Huafei SONG. Few-shot object detection on Thangka via multi-scale context information[J]. Optics and Precision Engineering, 2023, 31(12): 1859

    Download Citation

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

    Category: Information Sciences

    Received: Aug. 22, 2022

    Accepted: --

    Published Online: Jul. 25, 2023

    The Author Email: Wenjin HU (wenjin_zhm@126.com)

    DOI:10.37188/OPE.20233112.1859

    Topics