• Laser & Optoelectronics Progress
  • Vol. 59, Issue 16, 1610011 (2022)
Pengfei Shang1、2, Yi Chen1、2、*, Weijia Lv1、2, Fang Zheng1、2, and Jielong Wang1、2
Author Affiliations
  • 1College of Surveying and Geo-Informatics, Tongji Univesity, Shanghai 200092, China
  • 2Key Laboratory of Modern Engineering Surveying of National Administration of Surveying, Mapping and Geoinformation, Shanghai 200092, China
  • show less
    DOI: 10.3788/LOP202259.1610011 Cite this Article Set citation alerts
    Pengfei Shang, Yi Chen, Weijia Lv, Fang Zheng, Jielong Wang. Point-Cloud Semantic Segmentation Network Considering Normals[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1610011 Copy Citation Text show less
    Framework of PointNet (vanilla)
    Fig. 1. Framework of PointNet (vanilla)
    Framework of PPCAN (modified from Ref. [7])
    Fig. 2. Framework of PPCAN (modified from Ref. [7])
    Comparison of accuracy in training epoch
    Fig. 3. Comparison of accuracy in training epoch
    Semantic segmentation results of column. (a) PointNet; (b) PPCAN; (c) ground truth
    Fig. 4. Semantic segmentation results of column. (a) PointNet; (b) PPCAN; (c) ground truth
    Semantic segmentation results of sofa. (a) PointNet; (b) PPCAN; (c) ground truth
    Fig. 5. Semantic segmentation results of sofa. (a) PointNet; (b) PPCAN; (c) ground truth
    Increment of IoU and accuracy of PPACAN compared with PointNet
    Fig. 6. Increment of IoU and accuracy of PPACAN compared with PointNet
    Semantic segmentation results of board. (a) PointNet; (b) PPCAN; (c) ground truth
    Fig. 7. Semantic segmentation results of board. (a) PointNet; (b) PPCAN; (c) ground truth
    Results of 8 times predication on both PointNet and PPCAN. (a) PointNet; (b) PPCAN
    Fig. 8. Results of 8 times predication on both PointNet and PPCAN. (a) PointNet; (b) PPCAN
    Error characteristics of IoU with PPCAN. (a) Range; (b) standard deviation
    Fig. 9. Error characteristics of IoU with PPCAN. (a) Range; (b) standard deviation
    ModelOAMAmIoU
    PointNet82.967.457.1
    PPCAN85.274.561.0
    Table 1. Results of semantic segmentation on S3DIS
    ModelCeilingFloorWallBeamColumnWindowDoorTableChairSofaBookcaseBoardClutter
    PointNet96.198.589.255.823.565.890.676.666.420.164.165.364.1
    PPCAN97.897.787.371.165.786.383.679.467.465.770.227.470.0
    Table 2. Comparison of accuracy among 13 classes
    ModelCeilingFloorWallBeamColumnWindowDoorTableChairSofaBookcaseBoardClutter
    PointNet91.696.069.147.522.461.270.064.055.910.348.955.350.7
    PPCAN93.997.073.762.847.457.369.567.558.230.153.226.655.3
    Table 3. Comparison of IoU among 13 classes
    Pengfei Shang, Yi Chen, Weijia Lv, Fang Zheng, Jielong Wang. Point-Cloud Semantic Segmentation Network Considering Normals[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1610011
    Download Citation