Parametric Surface Constrained Upsampler Network for Point Cloud

Designing a point cloud upsampler, which aims to generate clean and dense given sparse representation, is fundamental challenging problem in computer vision. A line of attempts achieves this goal by establishing point-to-point mapping function via deep neural networks. However, these approaches are prone produce outlier points due the lack explicit surface-level constraints. To solve problem, we introduce novel surface regularizer into upsampler network forcing learn underlying parametric represented bicubic functions rotation functions, where new generated then constrained on surface. These designs integrated two different networks for tasks that take advantages upsampling layers -- completion evaluation. The state-of-the-art experimental results both demonstrate effectiveness proposed method. implementation code will be available at https://github.com/corecai163/PSCU.