It is a deep learning architecture specifically designed for 3D object detection from point clouds.
The name “PointPillars” comes from how it processes data: instead of treating the point cloud as random 3D points or complex 3D voxels, it organizes into vertical columns called “pillars”.