Scale-Aware Attention-Based PillarsNet (SAPN) Based 3D Object Detection for Point Cloud

Table 2

BEV detection performance: average precision (AP) for car objects in easy, moderate, and hard splits of the KITTI validation set.

Method	Modality	Easy	Mod	Hard
MV3D [4]	RGB + LiDAR	86.55	78.10	76.67
F-PointNet [16]	RGB + LiDAR	88.16	84.02	76.44
AVOD-FPN [15]	RGB + LiDAR	84.41	74.44	68.65
VoxelNet [5]	LiDAR	89.60	84.81	78.57
SECOND [14]	LiDAR	89.96	87.07	79.66
PointPillars [6]	LiDAR	89.79	86.60	84.17
Ours	LiDAR	91.18	88.37	85.32