Density-Aware Superpoint Grouping with a Deep Residual MLP for 3D Point Cloud Segmentation

Three-dimensional (3D) point cloud segmentation plays a crucial role in computer vision and 3D scene understanding. This paper proposes an efficient segmentation framework that integrates density-aware preprocessing, superpoint construction, and classification using a deep residual multilayer perceptron (MLP). First, the raw point cloud is pre-segmented into regions with locally consistent densities, and these regions are then grouped into superpoints to capture meaningful semantic structures. Subsequently, a deep residual MLP classifier is designed to learn the features of these superpoints and predict their labels, thereby achieving semantic segmentation of the point cloud. Experiments conducted on the ShapeNet dataset demonstrate that the proposed method achieves competitive segmentation accuracy while maintaining low computational cost, validating the effectiveness of the density-aware superpoint construction and deep residual MLP classification strategy.