基于边界点估计与稀疏卷积神经网络的三维点云语义分割
杨军,张琛

Semantic segmentation of 3D point cloud based on boundary point estimation and sparse convolution neural network
Jun YANG,Chen ZHANG
表 2 不同方法在SemanticKITTI数据集上的分割精度对比
Tab.2 Comparison of segmentation accuracy of different methods on SemanticKITTI dataset
方法mIoU/%IoU/%
roadside-
walk
par-
king
other-groundbuil-
ding
cartruckbicy-
cle
motor-
cycle
other-vehiclevegeta-
tion
trunkterrainper-
son
bicy-
clist
motor-
cyclist
fencepoletraffic-sign
PointNet [5]14.661.635.715.81.441.446.30.11.30.30.831.04.617.60.20.20.012.92.43.7
SPG [30]17.445.028.51.60.664.349.30.10.20.20.848.927.224.60.32.70.120.815.90.8
PointNet++ [6]20.172.041.818.75.662.353.70.91.90.20.246.513.830.00.91.00.016.96.08.9
TangentConv [28]40.983.963.933.415.483.490.815.22.716.512.179.549.358.123.028.48.149.035.828.5
SpSequenceNet [38]43.190.173.957.627.191.288.529.224.00.022.784.066.065.76.30.00.067.750.848.7
HPGCNN [39]50.589.573.658.834.691.293.121.06.517.623.384.465.970.032.130.014.765.545.541.5
RangeNet++ [40]52.291.875.265.027.887.491.425.725.734.423.080.555.164.638.338.84.858.647.955.9
RandLA-Net [18]53.990.773.760.320.486.994.240.126.025.838.981.461.366.849.248.27.256.349.247.7
PolarNet [41]54.390.874.461.721.790.093.822.940.330.128.584.065.567.843.240.25.661.351.857.5
3D-MiniNet [42]55.891.674.564.225.489.490.528.542.342.129.482.860.866.747.844.114.560.848.056.6
SAFFGCNN [43]56.689.973.963.535.191.595.038.333.235.128.784.467.169.545.343.57.366.154.353.7
KPConv [36]58.888.872.761.331.690.596.033.430.242.531.684.869.269.161.561.611.864.256.448.4
BAAF-Net [35]59.990.974.462.223.689.895.448.731.835.546.782.763.467.949.555.753.060.853.752.0
TORNADONet [44]61.190.875.365.327.589.693.143.153.044.439.484.164.369.661.656.720.262.955.064.2
FusionNet [20]61.391.877.168.830.892.595.341.847.537.734.584.569.868.559.556.811.969.460.066.5
本研究方法62.792.778.571.631.591.495.540.946.148.042.285.268.470.263.954.323.868.656.762.8