Large-scale point cloud semantic segmentation via local perception and global descriptor vector

Ziyin Zeng, Yongyang Xu, Zhong Xie, Wei Tang, Jie Wan, Weichao Wu

Abstract: Highlights•LAFA module captures local context to address local perception ambiguity.•Aggregation loss further mitigates the local perceptual ambiguity.•C-VLAD module captures global context to address feature insufficient.•Experiments demonstrate effectiveness of the proposed method.