Definition
PointNet++ is a model for processing 3D point cloud data. The model addresses some of the limitations of its predecessor PointNet by introducing hierarchical feature learning and multiscale grouping.
Architecture
The multi-scale grouping and multi-resolution grouping
Grouping
Constructs a graph in which nearby points are connected. Typically, this is either done via k -nearest neighbor search or via ball queries.
Neighborhood Aggregation
For each point, executes a Graph Neural Network layer that. It allows PointNet++ to capture local context at different scales.