Semantic Segmentation for Point Clouds via Semantic-Based Local Aggregation and Multi-Scale Global Pyramid

Recently, point-based networks have begun to prevail because they retain more original geometric information from point clouds than other deep learning-based methods. However, we observe that: (1) the set abstraction design for local aggregation in neglects that points a region may belong different semantic categories, and (2) most works focus on single-scale features while ignoring importance of multi-scale global features. To tackle above issues, propose two novel strategies named semantic-based (SLA) pyramid (MGP). The key idea SLA is augment based similarity neighboring region. Additionally, hierarchical (HGA) module extend feature aggregation. Based HGA, introduce MGP obtain discriminative multi-resolution cloud scenes. Extensive experiments prevailing benchmarks, S3DIS Semantic3D, demonstrate effectiveness our method.