Optics and Precision Engineering, Volume. 31, Issue 6, 950(2023)

Lightweight vehicle detection using long-distance dependence and multi-scale representation

Xiuping JING and Ying TIAN*
Author Affiliations
  • College of Computer Science and Software Engineering,University of Science and Technology Liaoning, Anshan114000,China
  • show less
    Figures & Tables(16)
    Sketch map of long distance dependence
    Structure of Visual Attention Network
    Structure of the Res2 module when feature maps subset are 4
    Structure of original C3 module
    Structure of Res2C3 and VANC3
    Structure of improved Yolov5s
    PR curve of the Pascal VOC and MS COCO vehicle dataset
    Different structure of Res2C3
    Inference result
    Anti occlusion ability test
    • Table 1. Different feature map subsets division methods

      View table
      View in Article

      Table 1. Different feature map subsets division methods

      Setting#Params(M)FLOPs(G)
      org7.317.1
      7w×8s7.216.2
      13w×4s7.316.6
      13w×6s7.618.6
    • Table 2. Experimental configuration and environment

      View table
      View in Article

      Table 2. Experimental configuration and environment

      实验环境具体配置信息
      系统Centos7
      Python版本3.8
      GPUGTX 1080ti×2
      Cuda版本10.1
      实验框架Pytorch 1.7.1
    • Table 3. Experimental results of each networks on Puscal VOC vehicle dataset

      View table
      View in Article

      Table 3. Experimental results of each networks on Puscal VOC vehicle dataset

      Model nameImg size#Params/MFLOPs/GmAP(%) Pascal VOCmAP(%) MS COCInference time/ms
      Yolov5s640×6407.317.182.562.36.1
      Yolov5s-improved640×6408.117.384.664.06.9
      Yolov4-tiny640×6406.417.470.354.33.9
      Yolov3-tiny640×6408.913.370.251.53.7
      Retinanet1 000×60036.2122.284.461.252.8
    • Table 4. Ablation Experiments on Pascal VOC and MS COCO vehicle dataset

      View table
      View in Article

      Table 4. Ablation Experiments on Pascal VOC and MS COCO vehicle dataset

      VANC3Res2C3

      FLOPs

      /G

      mAP/%

      VOC

      mAP

      /%

      COCO

      Inference time/ms
      ××17.182.562.36.1
      ×17.783.262.86.2
      17.384.664.06.9
    • Table 5. Influence of VANC3 for mAP on different stages

      View table
      View in Article

      Table 5. Influence of VANC3 for mAP on different stages

      Stage1Stage2Stage3Stage4mAP/%
      ×××82.8
      ×××82.0
      ×××81.7
      ×××83.2
    • Table 6. Influence of Res2C3 residual module on mAP for more different structures

      View table
      View in Article

      Table 6. Influence of Res2C3 residual module on mAP for more different structures

      Model NameFlops/GmAP/%Inference time/ms
      org17.182.56.1
      add K117.984.27.0
      13w×6s19.284.48.2
    Tools

    Get Citation

    Copy Citation Text

    Xiuping JING, Ying TIAN. Lightweight vehicle detection using long-distance dependence and multi-scale representation[J]. Optics and Precision Engineering, 2023, 31(6): 950

    Download Citation

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

    Category: Information Sciences

    Received: Jun. 9, 2022

    Accepted: --

    Published Online: Apr. 4, 2023

    The Author Email: TIAN Ying (astianying@126.com)

    DOI:10.37188/OPE.20233106.0950

    Topics