A Local-Global Feature Fusing Method for Point Clouds Semantic Segmentation

In recent years, the abundance of information in 3D data has made semantic segmentation point clouds a topic great interest. However, current methods often rely solely on original three-dimensional coordinates cloud as input geometric features, leading to poor generalization performance. Additionally, occlusion can negatively impact accuracy when only local is considered. To address these issues, this paper proposes network named LGFF-Net. fully utilize clouds, we designed Local Feature Aggregation (LFA) module that treats and equally preserves properties while cross-augmenting them. On other hand, proposed simple effective Global Extraction (GFE) extract global features. Finally, hierarchically fuse features using U-shaped structure. Compared state-of-the-art networks, our method achieves competitive results several benchmark datasets, including Semantic Topographic Point Labeling-Synthetic 3D, Toronto_3D, Stanford Large 3-D Indoor Space, ScanNet. We also conduct multiple ablation experiments validate efficacy