A Multibranch Object Detection Method for Traffic Scenes

Table 10

The average precision (AP) of different methods on KITTI dataset.

Model	Car			Cyclist			Pedestrian
	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
RCNN [4]	44.27	35.49	21.78	30.34	22.17	15.68	41.24	33.55	25.57
Faster RCNN [9]	52.14	41.23	30.77	34.54	25.24	18.29	39.67	26.54	18.23
SSD [15]	83.55	67.87	50.27	57.17	42.14	35.23	62.19	44.53	35.78
YOLOv3 [17]	87.22	71.28	64.67	72.13	60.06	42.77	77.32	65.34	55.58
Mask RCNN [13]	84.39	68.28	58.89	73.68	58.45	40.08	78.32	63.69	50.21
SINet [2]	88.35	77.49	62.57	75.72	60.29	43.12	80.49	65.97	54.68
MBNet	88.67	74.44	65.98	74.53	62.65	45.30	82.59	66.22	56.21