Abstract
Surface-based geodesic topology provides strong cues for object semanticanalysis and geometric modeling. However, such connectivity information is lostin point clouds. Thus we introduce GeoNet, the first deep learning architecturetrained to model the intrinsic structure of surfaces represented as pointclouds. To demonstrate the applicability of learned geodesic-awarerepresentations, we propose fusion schemes which use GeoNet in conjunction withother baseline or backbone networks, such as PU-Net and PointNet++, fordown-stream point cloud analysis. Our method improves the state-of-the-art onmultiple representative tasks that can benefit from understandings of theunderlying surface topology, including point upsampling, normal estimation,mesh reconstruction and non-rigid shape classification.