Feature Based Sampling: A Fast and Robust Sampling Method for Tasks Using 3D Point Cloud

Point cloud data sets are frequently used in machines to sense the real world because sensors such as LIDAR readily available be many applications including autonomous cars and drones. PointNet PointNet++ widely point-wise embedding methods for interpreting clouds. However, even recent models based on PointNet, real-time inference is still challenging. The solution a faster sampling, where, sampling method reduce number of points that computed next module. Furthest Sampling (FPS) used, but disadvantage it slow difficult select critical points. In this paper, we introduce Feature-Based (FBS), novel applies attention technique. results show significant speedup training time while accuracy similar previous methods. Further experiments demonstrate proposed better suited preserve or discard unimportant