Details

Autor(en) / Beteiligte

• Shi, Shaoshuai
• Jiang, Li
• Deng, Jiajun
• Wang, Zhe
• Guo, Chaoxu
• Shi, Jianping
• Wang, Xiaogang
• Li, Hongsheng

Titel

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

Ist Teil von

• International journal of computer vision, 2023-02, Vol.131 (2), p.531-551

Ort / Verlag

New York: Springer US

Erscheinungsjahr

2023

Link zum Volltext

Quelle

SpringerLink

Beschreibungen/Notizen

• 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 × 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 150 m × 150 m .