The demand for 3D modeling using LiDAR as the primary source for observing, planning, and managing urban areas has increased. Using LiDAR data improves the accuracy of the modeling so that it can be used for policy determination and infrastructure planning. Various kinds of research on LiDAR data have been carried out, one of which is indoor and outdoor LiDAR segmentation. For outdoor cases, LiDAR data can be obtained from two points of view, namely ground view and aerial view. In this paper, we discuss the advancements and challenges of LiDAR 3D modeling in building segmentation that we have carried out. We collect LiDAR data with unmanned aerial vehicles. We use several algorithms such as PointNet and the Dynamic Graph Convolutional Neural Network variations to group structures from LiDAR data. The result is that the proposed method can segment buildings, surfaces, and vegetation well. The average accuracy produced for the Kupang and Depok datasets reaches 70%-80%.