Optics and Precision Engineering, Volume. 32, Issue 18, 2823(2024)

3D point cloud classification and segmentation based on dual attention and weighted dynamic graph convolution

Jian XIAO... Xiaohong WANG*, Wei LI, Yifei YANG and Ji LUO |Show fewer author(s)
Author Affiliations
  • School of Mining, Guizhou University, Guiyang550000, China
  • show less
    Figures & Tables(18)
    Point cloud classification and segmentation networks based on DAM and WEM
    Composition of WEM
    Weighted KNN module
    ECC module
    Dual attention module
    Training epochs and overall accuracy curve graph
    Visualization results of part segmentation
    Comparison of the impact of K on the performance of classification network
    Comparison of the impact of N on the performance of classification network
    Robustness analysis experiment results
    Comparison of the impact of K on the performance of segmentation network
    • Table 1. Main parameter settings of the experiment

      View table
      View in Article

      Table 1. Main parameter settings of the experiment

      DatasetPointsOptimizerLearning rateCosine AnnealingBatch sizeTest batch sizeEpochs
      ModelNet401 024AdamW0.0011×10-5168250
      ShapeNet Part2 048AdamW0.0011.5×10-584200
    • Table 2. Comparison of 3d point cloud classification results on modelnet40 dataset

      View table
      View in Article

      Table 2. Comparison of 3d point cloud classification results on modelnet40 dataset

      MethodsInput TypeOA/%mAcc/%
      MVCNN1Image90.1
      VoxNet2Voxel85.9
      PointNet5Points(1k)89.286.2
      PointNet++6Points(5k)+Normal91.789.5
      PointWeb11Points(1k)+Normal92.389.4
      Point-Transformer21Points(1k)93.7
      PointConv23Points(1k)+Normal92.4
      PointASNL24Points(1k)+Normal93.2
      SpiderCNN25Points(1k)+Normal92.4
      GCN3D20Points(0.768k)93.090.3
      RF-DGCNN22Points(1k)93.6290.86
      KPConv9Points(1k)92.9
      DGCNN7Points(1k)92.990.2
      Linked-DGCNN12Points(3k)92.689.7
      OursPoints(1k)93.3690.74
    • Table 3. Comparison of the results for the parameters and forward time of different classification networks

      View table
      View in Article

      Table 3. Comparison of the results for the parameters and forward time of different classification networks

      MethodsParameters/MForward Time/msOA/%
      PointNet53.483.289.2
      PointNet++61.484.891.7
      DGCNN71.846.392.9
      Point-Transformer211.73136.493.7
      Ours2.088.393.36
    • Table 4. Comparison of 3d point cloud segmentation results on shapenet part dataset

      View table
      View in Article

      Table 4. Comparison of 3d point cloud segmentation results on shapenet part dataset

      MethodsmIoU/%ariplaneBagCapCarChairEarphoneGuitarKnife
      PointNet583.783.478.782.574.989.673.091.585.9
      PointNet++685.182.479.087.777.390.871.891.085.9
      PointConv2185.7
      PointASNL2286.184.184.787.979.792.273.791.087.2
      GCN3D2485.283.378.883.277.590.875.890.886.0
      DTNet2685.683.081.484.378.490.974.391.087.3
      3D-GCN2785.183.184.086.677.590.374.190.986.4
      DGCNN785.284.083.486.777.890.674.791.287.5
      Linked-DGCNN1284.785.488.690.890.988.6
      Ours85.9686.586.689.480.491.781.391.988.4
    • Table 4. Comparison of 3d point cloud segmentation results on shapenet part dataset

      View table
      View in Article

      Table 4. Comparison of 3d point cloud segmentation results on shapenet part dataset

      MethodsLampLaptopMotorbikeMugPistolRocketSkateboardTable
      PointNet580.895.365.293.081.257.972.880.6
      PointNet++683.795.371.694.181.358.776.482.6
      PointConv21
      PointASNL2284.295.874.495.281.063.076.383.2
      GCN3D2483.895.266.493.481.351.267.183.0
      DTNet2684.795.669.094.482.559.076.483.5
      3D-GCN2783.895.666.894.881.359.675.782.8
      DGCNN782.895.766.394.981.163.574.582.6
      Linked-DGCNN1282.495.474.781.9
      Ours84.596.472.996.085.463.678.882.9
    • Table 5. Validation of the effectiveness of each module in point cloud classification task

      View table
      View in Article

      Table 5. Validation of the effectiveness of each module in point cloud classification task

      NetsDAM

      ECC/

      KNN/WKNN

      TopKOA/ %mAcc/ %
      A×KNN×91.4587.68
      BKNN×92.1088.82
      CECC+KNN×92.1888.83
      DECC+KNN92.3089.31
      EWKNN×92.4289.25
      FWKNN92.6388.67
      GECC+WKNN×92.9189.89
      HECC+WKNN93.3690.74
    • Table 6. Validation of the effectiveness of each module in point cloud segmentation task

      View table
      View in Article

      Table 6. Validation of the effectiveness of each module in point cloud segmentation task

      DAMECC+WKNN/KNNOA/%mIoU/%
      ×KNN94.0384.65
      KNN94.1384.94
      ECC+KNN94.2885.29
      WKNN94.2585.32
      ECC+WKNN94.5685.96
    Tools

    Get Citation

    Copy Citation Text

    Jian XIAO, Xiaohong WANG, Wei LI, Yifei YANG, Ji LUO. 3D point cloud classification and segmentation based on dual attention and weighted dynamic graph convolution[J]. Optics and Precision Engineering, 2024, 32(18): 2823

    Download Citation

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

    Category:

    Received: Jan. 25, 2024

    Accepted: --

    Published Online: Nov. 18, 2024

    The Author Email: WANG Xiaohong (xhwang@gzu.edu.cn)

    DOI:10.37188/OPE.20243218.2823

    Topics