Learning Hyper Surfaces -- Supplementary Material -- Qing Li

For Eq.(5) in the paper, we provide a more detailed derivation process to clarify its relationship to the The red points are weighted higher than the blue points. It includes missing data and noise patterns that are very different from the PCPNet and SceneNN dataset. Note that there is no reconstructed mesh data or ground-truth normal for the Semantic3D dataset. The ground-truth normal of this dataset is not available. We report the number of the learnable network parameters of each model and its execution time on a single NVIDIA 2080 Ti GPU.