Pointfilter: Point Cloud Filtering via Encoder-Decoder Modeling
share
Point cloud filtering is a fundamental problem in geometry modeling and processing. Despite of advancement in recent years, the existing methods still suffer from two issues: they are either designed without preserving sharp features or less robust in preserving geometric features; they usually have many parameters and require tedious parameter tuning. In this paper, we propose a novel deep learning approach that automatically and robustly filters point clouds with removing noise and preserving sharp features and geometric details. Our point-wise learning architecture consists of an encoder and a decoder. The encoder directly takes points (a point and its neighbors) as input, and learns a latent representation vector which is gone through the decoder and related to the ground-truth position. Our trained network can automatically infer a corresponding quality point set to a noisy point cloud input. Extensive evaluations show that our approach outperforms the state-of-the-art deep learning techniques in terms of visual quality and error metrics. We will make our code and dataset publicly available.
READ FULL TEXT