A Depth Image Fusion Network for 3D Point Cloud Semantic Segmentation

It is widely known that the 3D point cloud is the most intuitive way to depict real scenes. Till now, some deep learning networks for point cloud data that have been proposed. In this paper, we propose a simple and intuitive network framework for 3D point cloud semantic segmentation in small-scale scenes. PointSIFT is improved by designing the depth image features extraction module. Through the module structure, constructing a simple mapping relationship between the two feature space. By fusing two features, we achieve state-of-the-art performance, moreover higher training efficiency. We also test our network to real workbench scenarios. And there is no performance degradation due to real scene noise.