Deep learning–based 3D reconstruction and completion from partial LiDAR scans that combines classical geometric methods (Poisson Surface Reconstruction, PCA normals) with deep learning (Sparse UNet) to predict missing parts of point clouds.