Key points are not available for this paper at this time.
3D point cloud semantic segmentation plays a critical role in interpreting intricate spatial environments. However, current methodologies often struggle to integrate channel-spatial interdependencies and lack sensitivity to boundary details, limiting precision at object edges. To address these challenges, this paper introduces CSEANet, a novel semantic segmentation network for enhanced point cloud analysis. Specifically, we propose the Channel Spatial Point Attention (CSPA) module to capture complex feature representations by synergistically modelling channel and spatial information. Furthermore, an Edge Information Awareness (EIA) module is introduced to preserve relevant boundary features while suppressing noise, thereby improving edge delineation . Experimental results on the ShapeNetPart and S3DIS datasets demonstrate competitive performance, particularly in boundary-sensitive tasks. To verify the robustness and generalizability of the model, we conducted extensive tests on outdoor large-scale datasets, including For-instance and SemanticKITTI. The results indicate that CSEANet effectively adapts to diverse and complex outdoor environments, outperforming several state-of-the-art baselines.
Liu et al. (Thu,) studied this question.