Learning modified indicator functions for surface reconstruction

Surface reconstruction is a fundamental problem in 3D graphics. In this paper, we propose learning-based approach for implicit surface from raw point clouds without normals. Our method inspired by Gauss Lemma potential energy theory, which gives an explicit integral formula the indicator functions. We design novel deep neural network to perform and learn modified functions un-oriented noisy clouds. concatenate features with different scales accurate point-wise contributions integral. Moreover, Element Feature Extractor local shape properties. Experiments show that our generates smooth surfaces high normal consistency noise achieves state-of-the-art performance compared current data-driven non-data-driven approaches.