Optics and Precision Engineering, Volume. 33, Issue 5, 777(2025)

Shape adaptive feature aggregation network for point cloud classification and segmentation

Zhihao JIANG1, Meixiang ZHANG1, Weitao XUE2, Lina FU1, Jing WEN1, Yongqiang LI2、*, and Hong HUANG1、*
Author Affiliations
  • 1Key Laboratory of Optoelectronic Technology and System, Ministry of Education, Chongqing University, Chongqing400044, China
  • 2Product Testing Center, Beijing Institute of Space Machinery and Electronics, Beijing100094, China
  • show less
    Figures & Tables(12)
    Overall structure of Shape Adaptive Neighbor Feature Aggregation Network(SANFA-Net)
    Specific structure of deep feature extraction layer
    Specific structure of feature decoding layer
    Examples of three point cloud datasets
    Experimental results of ablation visualization
    Visualization experiments on ShapeNet dataset
    Visualization experiments on S3dis dataset
    • Table 1. Experimental hyperparameter configuration

      View table
      View in Article

      Table 1. Experimental hyperparameter configuration

      ParameterModelNet40ShapeNetS3dis
      Input points1 0242 0484 096
      Batchsize241212
      Epochs40020064
      Learning rate0.0010.0010.01
      Lr_schedulerCosine Annealing LR
      OptimizerAdamW
      Encoding layers334
      Decoding layers-34
    • Table 2. Ablation experiments of SANFA-Net on three tasks

      View table
      View in Article

      Table 2. Ablation experiments of SANFA-Net on three tasks

      ModelModulesModelNet40ShapeNetS3dis
      baselineASDMALFEMIns.accCls.accCls.mIoUIns.mIoUIns.accIns.mIoU
      (a)92.5389.1782.1685.2162.0353.53
      (b)93.3890.6382.8385.5965.1857.30
      (c)93.9291.4183.3585.8667.2859.68
    • Table 3. Comparative experiments of SANFA-Net on ModelNet40 dataset

      View table
      View in Article

      Table 3. Comparative experiments of SANFA-Net on ModelNet40 dataset

      MethodPara(M)Ins.acc/%
      PointConv22-92.5
      PointConT23-93.5
      3D-GCN160.8992.1
      DGCNN181.8192.2
      DeepGCNs172.293.6
      PointNet143.589.2
      PointNet++151.4892.5
      CSANet2413.292.8
      PCT-KF25-93.5
      Ours1.7293.9
    • Table 4. Comparative experiments of SANFA-Net on ShapeNet dataset

      View table
      View in Article

      Table 4. Comparative experiments of SANFA-Net on ShapeNet dataset

      MethodPara/MCls.mIoU/%Ins.mIoU/%
      ConvPoint26-83.485.8
      PointConv22-82.885.7
      DGCNN181.4682.385.1
      3D-GCN161.7182.785.3
      PointNet148.3480.483.7
      PointNet++151.4181.985.1
      DPFANet27--85.5
      PCT-KF25--85.6
      CSANet2415.45-85.7
      Ours1.6583.485.9
    • Table 5. Comparative experiments of SANFA-Net on S3dis dataset

      View table
      View in Article

      Table 5. Comparative experiments of SANFA-Net on S3dis dataset

      MethodsPara/MIns.acc/%Ins.mIoU/%
      PointCNN2811.7463.957.3
      PCCN29-67.058.3
      DeepGCNs17--53.6
      3D-GCN161.71-51.9
      DGCNN180.9861.552.0
      PointNet143.5349.041.1
      PointNet++150.9762.053.5
      CSANet2415.45-53.3
      DPFANet27--55.2
      Ours2.0167.359.7
    Tools

    Get Citation

    Copy Citation Text

    Zhihao JIANG, Meixiang ZHANG, Weitao XUE, Lina FU, Jing WEN, Yongqiang LI, Hong HUANG. Shape adaptive feature aggregation network for point cloud classification and segmentation[J]. Optics and Precision Engineering, 2025, 33(5): 777

    Download Citation

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

    Category:

    Received: Sep. 23, 2024

    Accepted: --

    Published Online: May. 20, 2025

    The Author Email: Yongqiang LI (hhuang@cqu.edu.cn)

    DOI:10.37188/OPE.20253305.0777

    Topics