Investigate Indistinguishable Points in Semantic Segmentation of 3D Point Cloud

This paper investigates the indistinguishable points (difficult to predict label) in semantic segmentation for large-scale 3D point clouds. The consist of those located complex boundary, with similar local textures but different categories, and isolate small hard areas, which largely harm performance segmentation. To address this challenge, we propose a novel Indistinguishable Area Focalization Network (IAF-Net), select adaptively by utilizing hierarchical features enhance fine-grained especially points. We also introduce multi-stage loss improve feature representation progressive way. Moreover, order analyze performances new evaluation metric called Points Based Metric (IPBM). Our IAF-Net achieves state-of-the-art on several popular datasets e.g. S3DIS ScanNet, clearly outperform other methods IPBM. code will be available at https://github.com/MingyeXu/IAF-Net.