Abstract: PointNet, which is the widely used point-wise embedding method and known as a universal approximator for continuous set functions, can process one million points per second. Nevertheless, ...

Stacked PFE architecture just as PointNet++ Pyramid PFE architecture Assume the first point feature encoder is deep enough. Use pyramid pooling to get multi-scale features. Unlike PointNet++ where ...

Abstract: Despite remarkable performance across a broad range of tasks, neural networks have been shown to be vulnerable to adversarial attacks. Many works focus on adversarial attacks and defenses on ...

Graph in the left shows the odometry prediction of the point-based model relative to ground truth in KITTI city 0. The graph in the right shows the performance for ICP for the same city.