Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 7, 980(2024)

Semantic segmentation method for street images with multi-dimensional features

Lei ZHU, Chenjie CHE*, Tongyu YAO, Yang PAN, and Bo ZHANG
Author Affiliations
  • College of Electronic Information,Xi'an Polytechnic University,Xi'an 710600,China
  • show less
    Figures & Tables(12)
    Overall architecture of the semantic partitioning network in this paper
    Structure of feature extraction sub-network
    Structure of target area enhancement module
    Dual attention module structure
    Feature pyramid grid structure
    Comparison of segmentation accuracy of different feature fusion modules
    Visualisation of the results of various types of target segmentation
    Visualisation of the results of different light divisions
    • Table 1. Partial training strategy parameters and corresponding details

      View table
      View in Article

      Table 1. Partial training strategy parameters and corresponding details

      训练策略参数细节
      迭代次数320k
      裁剪尺寸512×1 024
      初始学习率1.25×10-3
      最小学习率1×10-4
      动量0.9
      优化器SGD
      BN策略SyncBN
    • Table 2. Accuracy comparison experiments of semantic segmentation methods

      View table
      View in Article

      Table 2. Accuracy comparison experiments of semantic segmentation methods

      MethodCrop sizeLR SchedulemIoU/%mPA/%
      DeepLabV3512×1 02480k72.3281.25
      PSPNet512×1 02480k70.6678.75
      EMANet512×1 024160k68.0979.49
      BiSeNetV21 024×1 02480k71.0773.13
      STDC2512×1 02480k72.3180.33
      SegFormer1 024×1 024160k76.6084.79
      PointRend512×1 024320k73.0978.16
      MDFNet512×1 024320k80.1186.32
    • Table 3. Selected category accuracy comparison experiments

      View table
      View in Article

      Table 3. Selected category accuracy comparison experiments

      MethodmIoU/%PersonRiderCarTruckBusTrainMcycleBicycle
      PSPNet70.6680.2758.4593.7070.3156.5542.0762.0375.58
      DeepLabV372.3278.9956.1993.9470.5773.7858.4857.9973.88
      SegFormer76.6080.6657.8794.5280.8782.1467.2363.3475.59
      MDFNet-5077.5181.7261.5194.6482.1883.8669.6764.5476.08
      MDFNet-10180.1183.0263.1095.2682.6290.2884.9567.3478.05
    • Table 4. Ablation experiments in various modules

      View table
      View in Article

      Table 4. Ablation experiments in various modules

      BackboneBaselineTAEFPGmIoU/%mPA/%
      Res-5073.0380.28
      75.4281.35
      73.1780.56
      77.5185.35
      Res-10173.0978.16
      77.2782.50
      79.5585.79
      80.1186.32
    Tools

    Get Citation

    Copy Citation Text

    Lei ZHU, Chenjie CHE, Tongyu YAO, Yang PAN, Bo ZHANG. Semantic segmentation method for street images with multi-dimensional features[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(7): 980

    Download Citation

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

    Category: Research Articles

    Received: Jun. 7, 2023

    Accepted: --

    Published Online: Jul. 23, 2024

    The Author Email: Chenjie CHE (che147890@163.com)

    DOI:10.37188/CJLCD.2023-0208

    Topics