Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 5, 656(2023)

Object detection method based on CIoU improved bounding box loss function

Xiong-biao LIU1,2, Xian-zhao YANG1,2、*, Yang CHEN1,2, and Shuai-tong ZHAO1,2
Author Affiliations
  • 1College of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430081,China
  • 2Engineering Research Center of Metallurgical Automation and Measurement Technology,Ministry of Education,Wuhan 430081,China
  • show less
    Figures & Tables(11)
    CIoU illustration
    Image of the derivative function of υ
    Image of the derivative function of υυ
    BCIOU illustration
    Comparison chart of accuracy change of loss function
    Trend graph of loss function value
    Comparison chart of image visualization results
    • Table 1. Algorithm 1

      View table
      View in Article

      Table 1. Algorithm 1

      Algorithm 1: Calculating BCIoU Loss:

      Input: Bounding box of Ground truth Bgt =              xgt,ygt,wgt,hgtInput: Bounding box of Ground truth Bp =xp,yp,wp,hpOutput: LBCIoU1: ifBgt0 and BP0 do 2: LBCIoU=1-IoU+ρ2(Bgt,Bp)c2+αυυ+      ρ12(Bgt,Bp)c12+βυ1υ

    • Table 2. Algorithm 2

      View table
      View in Article

      Table 2. Algorithm 2

      Algorithm 2: Calculating BCIoU Loss:

      Input: Bounding box of Ground truth Bgt =              xgt,ygt,wgt,hgtInput: Bounding box of Ground truth Bp =              xp,yp,wp,hpOutput: LBCIoU1: ifBgt0 and BP0 do2: if BgtBP== 0 then3:LBCIoU=1-IoU+ρ2(Bgt,Bp)c2+αυυ4:else if BgtBp== Bgt or BgtBp== Bp5:LBCIoU=1-IOU+ρ2(Bgt,Bp)c2+αυυ+βυ1υ6:else7:LBCIoU=1-IoU+ρ2(Bgt,Bp)c2+αυυ+    ρ12(Bgt,Bp)c12+βυ1υ

    • Table 3. Results of ablation experiments

      View table
      View in Article

      Table 3. Results of ablation experiments

      Model\EvaluationAP50AP75APIteration
      YOLOv3+CIoU59.8927.6030.6590
      YOLOv3+CIoU[D]60.3229.4531.3684
      YOLOv3+CIoU[A]60.2229.1331.1287
      YOLOv3+CIoU[V]60.3929.7131.5688
    • Table 4. Test results of each loss function on PASCAL VOC 2007

      View table
      View in Article

      Table 4. Test results of each loss function on PASCAL VOC 2007

      LossEvaluation
      AP50AP55AP60AP65AP70AP75AP80AP85AP90AP95AP
      IoU59.6256.8752.8746.8637.9126.7015.235.130.940.0030.21
      DIoU59.4457.0053.0947.1638.1926.9915.405.560.890.0230.37
      Relative Improve/%-0.300.230.420.640.741.091.128.38-5.320.000.53
      CIoU59.8957.3953.3947.7239.1527.6015.565.430.880.0130.65
      Relative improve/%0.450.910.981.843.273.372.175.85-6.380.001.46
      Algorithm 160.6458.5154.3448.5741.3130.6418.376.801.190.0332.04
      Relative improve/%1.712.882.783.658.9714.7620.6232.5526.590.006.06
      Algorithm 260.8758.7454.7648.8641.6731.0418.666.981.230.0332.29
      Relative improve/%2.093.293.573.269.9116.2522.5236.0630.850.006.88
    Tools

    Get Citation

    Copy Citation Text

    Xiong-biao LIU, Xian-zhao YANG, Yang CHEN, Shuai-tong ZHAO. Object detection method based on CIoU improved bounding box loss function[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(5): 656

    Download Citation

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

    Category: Research Articles

    Received: Aug. 25, 2022

    Accepted: --

    Published Online: Jul. 4, 2023

    The Author Email: Xian-zhao YANG (yangxianzhao@wust.edu.cn)

    DOI:10.37188/CJLCD.2022-0282

    Topics