STD: Sparse-to-Dense 3D Object Detector for Point Cloud

ICCV 2019  ยท  Zetong Yang, Yanan sun, Shu Liu, Xiaoyong Shen, Jiaya Jia ยท

We present a new two-stage 3D object detection framework, named sparse-to-dense 3D Object Detector (STD). The first stage is a bottom-up proposal generation network that uses raw point cloud as input to generate accurate proposals by seeding each point with a new spherical anchor. It achieves a high recall with less computation compared with prior works. Then, PointsPool is applied for generating proposal features by transforming their interior point features from sparse expression to compact representation, which saves even more computation time. In box prediction, which is the second stage, we implement a parallel intersection-over-union (IoU) branch to increase awareness of localization accuracy, resulting in further improved performance. We conduct experiments on KITTI dataset, and evaluate our method in terms of 3D object and Bird's Eye View (BEV) detection. Our method outperforms other state-of-the-arts by a large margin, especially on the hard set, with inference speed more than 10 FPS.

PDF Abstract ICCV 2019 PDF ICCV 2019 Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
3D Object Detection KITTI Cars Easy STD AP 86.61% # 14
Birds Eye View Object Detection KITTI Cars Easy STD AP 89.66 # 5
Birds Eye View Object Detection KITTI Cars Hard STD AP 86.89 # 1
3D Object Detection KITTI Cars Hard STD AP 76.06% # 8
3D Object Detection KITTI Cars Moderate STD AP 77.63% # 17
Birds Eye View Object Detection KITTI Cars Moderate STD AP 87.76% # 4
Birds Eye View Object Detection KITTI Cyclists Easy STD AP 81.04 # 2
3D Object Detection KITTI Cyclists Easy STD AP 78.89% # 5
Birds Eye View Object Detection KITTI Cyclists Hard STD AP 57.85 # 2
3D Object Detection KITTI Cyclists Hard STD AP 55.77% # 6
3D Object Detection KITTI Cyclists Moderate STD AP 62.53% # 6
Birds Eye View Object Detection KITTI Cyclists Moderate STD AP 65.32% # 2
3D Object Detection KITTI Pedestrians Easy STD AP 53.08% # 3
Birds Eye View Object Detection KITTI Pedestrians Easy STD AP 60.99 # 1
3D Object Detection KITTI Pedestrians Hard STD AP 41.97% # 3
Birds Eye View Object Detection KITTI Pedestrians Hard STD AP 45.89 # 2
3D Object Detection KITTI Pedestrians Moderate STD AP 44.24% # 5
Birds Eye View Object Detection KITTI Pedestrians Moderate STD AP 51.39% # 2

Methods