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

Table 4

Average precisions of BEV detection on the KITTI validation split. , , and represent average precisions of easy, moderate, and hard splits, respectively.

Multi-resolution	Spatial-attention	SE-attention
✓	✕	✕	89.98	87.02	84.01
✕	✓	✕	90.02	87.68	84.08
✕	✕	✓	89.67	86.88	83.31
✓	✓	✕	91.03	87.91	84.91
✓	✕	✓	90.55	87.55	83.55
✕	✓	✓	90.32	88.02	84.12
✓	✓	✓	91.18	88.37	85.32