Fine-Grained Point Cloud Semantic Segmentation of Complex Railway Bridge Scenes from UAVs Using Improved DGCNN

Table 4

Segmentation results of different networks in scene (b).

Methods	Test mIoU (%)	Test bACC (%)	Parameters size (MB)
PointNet++ [23]	88.25	93.17	7.18
DGCNN [25]	88.96	93.67	2.45
KPConv [20]	89.36	94.04	14.10
Point transformer [35]	90.92	95.53	7.80
Swin3D-L [36]	83.87	89.68	60.75
Improved DGCNN	90.64	95.31	2.12