This repository implements the PointRCNN model from CVPR 2019, which uses a bottom-up approach to generate 3D object proposals from raw point clouds. The first stage produces proposals in a canonical coordinate system using a bin-based regression loss, and the second stage refines these proposals. It leverages PointNet++ for point cloud feature extraction and achieves state-of-the-art results on the KITTI 3D object detection benchmark.