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

Figure 7

Illustrations of feature maps with (a) and (b) without spatial-attention operations. In each subfigure, correct activations belong to car objects being labeled with red ellipses while wrong activations caused by complex backgrounds are labeled with yellow ellipse (best viewed in color).

(a)

(b)