Acta Optica Sinica, Volume. 45, Issue 8, 0815003(2025)

Contour Feature-Enhanced 3D Perception Method for Atypical Semantic Object

Farong Kou1、*, Kan Wang2, Yajun Zhao2, Tianxiang Yang2, and Gengyi Lü2
Author Affiliations
  • 1Institute of Electrical and Control Engineering, Xi’an University of Science and Technology, Xi’an 710054, Shaanxi , China
  • 2College of Mechanical Engineering, Xi’an University of Science and Technology, Xi’an 710054, Shaanxi , China
  • show less
    Figures & Tables(14)
    Structural diagram of SCE-FBOcc
    Contour convolution module
    Framework of contour encoding attention module
    Cross-scale semantic contour attention module
    Contour comparison diagram
    Framework of contour semantic assisted learning module
    Visualization results of Occupancy
    Visualization comparison of depth features
    Local comparison chart
    Comparison chart of occupancy grids
    Comparison Chart of Depth Estimation
    • Table 1. Campare with other method

      View table

      Table 1. Campare with other method

      ClassBEVDetBEVFormer-OccCTFOccFBOcc*SCE-BEVDetSCE-FBOcc
      Barrier15.2937.8739.9242.8515.6144.83
      Cons. vehicle1.357.3516.9621.132.3423.36
      Manmade14.5120.6220.6236.1715.7535.96
      Traffic cone0.1321.8122.7225.730.1827.64
      Pedestrian0.4321.6322.725.780.7327.94
      Car13.5742.4342.2446.9814.8747.94
      Truck7.4730.731.1335.798.6535.17
      Bus5.1840.4438.3944.265.4242.13
      Trailer4.5822.3822.9831.115.1334.44
      Bicycle4.1217.9820.5626.734.3228.55
      Motorcycle0.224.0824.5226.170.3328.56
      Dri. sur52.7655.3253.4278.2353.2379.41
      Vegetation15.2517.7118.134.4116.4735.98
      mIoU10.3727.7128.7836.5611.7737.83
    • Table 2. Ablation study

      View table

      Table 2. Ablation study

      ModelBarrierCons. vehicleManmadeTraffic coneDri. surmIoU
      No42.8521.1336.1725.7378.2340.82
      CEAM43.5122.2335.1225.9178.0240.95
      CEAM+SCALM43.2422.3135.8925.8379.2341.30
      CEAM+SCALM+CSCAM44.8323.3635.9627.6479.4142.24
    • Table 3. Interfering latency study

      View table

      Table 3. Interfering latency study

      ModuleParams /106Model params /106FLOPS /109Latency /msOverall latency /ms
      FPN+CM DepthNet27.3267.80105.935.6697.21
      CADN (PyTorch)31.8572.33119.4112.22103.75
      CADN (CUDA Ops)31.8572.33119.419.12100.53
    Tools

    Get Citation

    Copy Citation Text

    Farong Kou, Kan Wang, Yajun Zhao, Tianxiang Yang, Gengyi Lü. Contour Feature-Enhanced 3D Perception Method for Atypical Semantic Object[J]. Acta Optica Sinica, 2025, 45(8): 0815003

    Download Citation

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

    Category: Machine Vision

    Received: Dec. 20, 2024

    Accepted: Feb. 28, 2025

    Published Online: Apr. 27, 2025

    The Author Email: Farong Kou (koufarong@xust.edu.cn)

    DOI:10.3788/AOS241916

    CSTR:32393.14.AOS241916

    Topics