D2T-Net: A dual-domain transformer network exploiting spatial and channel dimensions for semantic segmentation of urban mobile laser scanning point clouds

Ziwei Luo, Ziyin Zeng, Jie Wan, Wei Tang, Zhongge Jin, Zhong Xie, Yongyang Xu

Abstract: Highlights•A novel network for semantic segmentation of urban MLS point clouds.•A dual domain Transformer-based framework, operated in both spatial and channel dimensions.•An enhanced self-attention mechanism to enrich spatial semantic cues.•A channel-wise transformer focuses on inter-channel relationships to enhance the global context.•A controlled flow gate is employed for selective information transfer.