Laser & Optoelectronics Progress, Volume. 62, Issue 2, 0212001(2025)

Multiview 3D Object Detection Based on Improved DETR3D

Yuhan Zhang1...2,3, Miaohua Huang1,2,3,*, Gengyao Chen1,2,3, Yanzhou Li1,2,3, and Yiming Wu1,23 |Show fewer author(s)
Author Affiliations
  • 1Hubei Key Laboratory of Advanced Technology for Automotive Components, Wuhan University of Technology, Wuhan 430070, Hubei , China
  • 2Hubei Collaborative Innovation Center for Automotive Components Technology, Wuhan University of Technology, Wuhan 430070, Hubei , China
  • 3Hubei Research Center for New Energy & Intelligent Connected Vehicle, Wuhan University of Technology, Wuhan 430070, Hubei , China
  • show less
    Figures & Tables(9)
    Algorithm overall framework diagram
    3D position encoding process
    Perceptual ability of queries for other queries from different distances
    Temporal feature sampling process
    Visualization of baseline model predictions
    Visualization of improved model predictions
    • Table 1. Comparison of results of different algorithms on the nuScenes validation set

      View table

      Table 1. Comparison of results of different algorithms on the nuScenes validation set

      AlgorithmModalityNDSmAPmATEmASEmAOEmAVEmAAEFPS
      DETR3DC0.4340.3490.7160.2680.3790.8420.2003.7
      BEVDetC0.4170.3490.6370.2690.4900.9140.26816.7
      PETRC0.4410.3660.7170.2670.4120.8340.1905.7
      BEVFormerC0.5170.4160.6730.2740.3720.3940.1983.0
      BEVDepthC0.5350.4120.5650.2660.3580.3310.1904.5
      Proposed algorithmC0.5080.4670.5640.4450.5180.4430.28915.6
    • Table 2. Comparison of results of different algorithms on the nuScenes test set

      View table

      Table 2. Comparison of results of different algorithms on the nuScenes test set

      AlgorithmModalityNDSmAPmATEmASEmAOEmAVEmAAE
      PointPillarL0.4530.305
      CenterPointL0.6550.580
      DETR3DC0.4790.4120.6410.2550.3940.8450.133
      BEVDetC0.4880.4240.5240.2420.3730.9500.148
      PETRC0.5040.4410.5930.2490.3830.8080.132
      BEVFormerC0.5690.4810.5820.2560.3750.3780.126
      BEVDepthC0.6090.5200.4450.2430.3520.3470.127
      Proposed algorithmC0.6140.5140.4620.2450.3430.2660.114
    • Table 3. Impact of different modules on model performance

      View table

      Table 3. Impact of different modules on model performance

      ModuleSettingNDSmAP
      Query formulation3D reference point0.4790.435
      Ours0.4890.443
      3DPENone0.4740.431
      Ours0.4890.443
      MSSAMHSA0.4670.403
      Ours0.4890.443
      TFFSN=4/8/160.459/0.475/0.4890.407/0.422/0.443
      T=1/4/80.401/0.462/0.4890.358/0.415/0.443
    Tools

    Get Citation

    Copy Citation Text

    Yuhan Zhang, Miaohua Huang, Gengyao Chen, Yanzhou Li, Yiming Wu. Multiview 3D Object Detection Based on Improved DETR3D[J]. Laser & Optoelectronics Progress, 2025, 62(2): 0212001

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Mar. 18, 2024

    Accepted: Apr. 30, 2024

    Published Online: Dec. 26, 2024

    The Author Email: Miaohua Huang (mh_huang@163.com)

    DOI:10.3788/LOP240912

    CSTR:32186.14.LOP240912

    Topics