PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection

31 Jan 2021  ·  Shaoshuai Shi, Li Jiang, Jiajun Deng, Zhe Wang, Chaoxu Guo, Jianping Shi, Xiaogang Wang, Hongsheng Li ·

3D object detection is receiving increasing attention from both industry and academia thanks to its wide applications in various fields. In this paper, we propose Point-Voxel Region-based Convolution Neural Networks (PV-RCNNs) for 3D object detection on point clouds. First, we propose a novel 3D detector, PV-RCNN, which boosts the 3D detection performance by deeply integrating the feature learning of both point-based set abstraction and voxel-based sparse convolution through two novel steps, i.e., the voxel-to-keypoint scene encoding and the keypoint-to-grid RoI feature abstraction. Second, we propose an advanced framework, PV-RCNN++, for more efficient and accurate 3D object detection. It consists of two major improvements: sectorized proposal-centric sampling for efficiently producing more representative keypoints, and VectorPool aggregation for better aggregating local point features with much less resource consumption. With these two strategies, our PV-RCNN++ is about $3\times$ faster than PV-RCNN, while also achieving better performance. The experiments demonstrate that our proposed PV-RCNN++ framework achieves state-of-the-art 3D detection performance on the large-scale and highly-competitive Waymo Open Dataset with 10 FPS inference speed on the detection range of 150m * 150m.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
3D Object Detection KITTI Cars Easy PV-RCNN++ AP 90.14% # 8
3D Object Detection KITTI Cars Easy val PV-RCNN++ AP 92.57 # 2
3D Object Detection KITTI Cars Hard PV-RCNN++ AP 77.15% # 3
3D Object Detection KITTI Cars Hard val PV-RCNN++ AP 82.69 # 2
3D Object Detection KITTI Cars Moderate PV-RCNN++ AP 81.88% # 8
3D Object Detection KITTI Cars Moderate val PV-RCNN++ AP 84.83 # 3

Methods