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

Table 2

Relevant parameters used in the proposed improved network.

Name	Variable	Value
Input feature		—
Block size		2 m
Epochs	—	600
Number of sampling points		4096
Batch size		32
Range of learning rates		()
Learning rate fine-tuning strategy	Warm-start cosine annealing, ,
Optimizer	[33]