Table 8 Ablation studies of the backbone...

Table 8

Ablation studies of the backbone structure adoption on the mod level of the KITTI val set and all the results are evaluated with average precision and calculated by 40 recall positions. Also, we provide FPS for the comparison of the model efficiency.

Methods	Car	3D Moderate Pedestrian	Cyclist	FPS
Voxel-RCNN, [6]	84.87	59.87	73.23	30
FocalConv^*, [4]	85.26	60.04	73.08	19
Ours	85.34	60.30	74.33	29

Note: The above results are reproduced by the publicly release model ([28]).

:Note that we only adopt the attention backbone structure while the remaining part is the same as the baseline model.

[ViewLarge]

Sharing Unavailable