• Optics and Precision Engineering
  • Vol. 29, Issue 10, 2504 (2021)
Jun Yang1,* and Min-min Zhang2
Author Affiliations
  • 1Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou730070, China
  • 2School of Automation and Electrical Engineering, Lanzhou Jiaotong University, Lanzhou730070, China
  • show less
    DOI: 10.37188/OPE.20212910.2504 Cite this Article
    Jun Yang, Min-min Zhang. Co-segmentation of three-dimensional shape clusters by shape similarity[J]. Optics and Precision Engineering, 2021, 29(10): 2504 Copy Citation Text show less
    Construction of spherical neighborhood
    Fig. 1. Construction of spherical neighborhood
    Convolution process of feature aggregation operator
    Fig. 2. Convolution process of feature aggregation operator
    Hierarchical co-segmentation network
    Fig. 3. Hierarchical co-segmentation network
    Segmentation results of point cloud shapes
    Fig. 4. Segmentation results of point cloud shapes

    Algorithms

    mIoU/%

    pIoU/%

    PointNet5

    80.4

    83.6

    PointASNL10]

    83.3

    86.1

    PointNet++19

    81.9

    85.1

    MACL27

    71.8

    75.2

    SPLATNet28

    83.7

    85.4

    Ours

    83.3

    86.0

    Table 1. Comparison of segmentation results between our algorithm and other algorithms
    Method

    Air.

    (2 690)

    Bag

    (76)

    Cap

    (55)

    Car

    (898)

    Chair

    (3 758)

    Ear.

    (69)

    Guitar

    (787)

    Knife

    (392)

    PointNet583.478.782.574.989.673.091.585.9
    PointASNL1084.184.787.979.792.273.791.087.2
    PointNet++1982.479.087.777.390.871.891.085.9
    MACL277574.574.054.581.366.588.484.4
    SPLATNet2883.284.389.180.390.775.592.187.1
    Ours83.284.085.780.391.178.590.388.4
    Method

    Lamp

    (1 547)

    Lap.

    (451)

    Motor.

    (202)

    Mug

    (184)

    Pistol

    (283)

    Rocket

    (66)

    Skate.

    (152)

    Table

    (5 271)

    PointNet580.895.365.293.081.257.972.880.6
    PointASNL1084.295.874.495.281.063.076.383.2
    PointNet++1983.795.371.694.181.358.776.482.6
    MACL2777.678.251.387.366.655.062.171.7
    SPLATNet2883.996.375.695.883.864.075.581.8
    Ours85.795.673.795.481.960.076.183.3
    Table 2. Comparison of IoU of different algorithms on various shapes
    #pointsInputTime
    Training/minTest/ms
    PointNet51 024Coordinates79322
    PointASNL102 048Coordinates--
    PointNet++191 024Coordinates+Normals1 26647
    MACL272 048Coordinates976.860.7
    SPLATNet28--94360.4
    Ours1 024Coordinates85941.9
    Table 3. Comparison of segmentation time between our algorithm and other algorithms
    SamplingpIoU/%
    Shared MLPFA Conv
    kNN67.585.2
    Random-PiS73.186.0
    Table 4. Effectiveness verification of different components for co-segmentation of 3D shape clusters
    Jun Yang, Min-min Zhang. Co-segmentation of three-dimensional shape clusters by shape similarity[J]. Optics and Precision Engineering, 2021, 29(10): 2504
    Download Citation