Table 1 Performance comparison with...

Table 1

Performance comparison with state-of-the-art models on the KITTI val set for pedestrian 3D detection in all three difficulties and also the comparison of the GPU memory usage (MiB) and the inference time (FPS). All the 3D detection results are evaluated with average precision and calculated by 40 recall positions.

Models	3D Pedestrian			Mem.	FPS
Models	Easy	Mod.	Hard	Mem.	FPS
PV-RCNN, [24]	62.72	54.51	49.87	8243	17
CT3D, [23]	61.04	55.55	51.05	6485	19
PDV, [12]	66.90	60.80	55.85	7302	21
Ours	68.53	61.01	56.41	5669	27

Models	3D Pedestrian	Mem.	FPS
PV-RCNN, [24]	62.72	54.51	49.87	8243	17
CT3D, [23]	61.04	55.55	51.05	6485	19
PDV, [12]	66.90	60.80	55.85	7302	21
Ours	68.53	61.01	56.41	5669	27

Note: Our model’s results are shown in bold, and the best results are underlined. Also, the above results are reproduced by the publicly release model ([28]).

[ViewLarge]

Sharing Unavailable