Local Feature Extraction Network for Point Cloud Analysis

Geometric feature extraction of 3D point clouds plays an important role in many computer vision applications such as region labeling, reconstruction, object segmentation, and recognition. However, hand-designed features on lack semantic information, so cannot meet these requirements. In this paper, we propose local network (LFE-Net) which focus extracting for analysis. Such geometric learning from a relation points can be used variety shape analysis problems classification, part matching. LFE-Net consists (LGR) module aims to learn high-dimensional express the between their neighbors. Benefiting additional singular values hierarchical neural networks, learned are robust permutation rigid transformation that they transformed into descriptors. Moreover, embed prior spatial information sub-features combining multiple levels. achieves state-of-the-art performances standard benchmarks including ModelNet40, ShapeNetPart.