Multiscale Feature Learning Based on Enhanced Feature Pyramid for Vehicle Detection

Table 1

Detection results of the proposed method and other methods on the KITTI testing set. The inference time is evaluated on single Nvidia RTX 3070 GPU per image.

Method	Average precision (%)			Inference speed (s)
	Easy	Moderate	Hard
Faster R-CNN [1]	86.71	81.84	71.12	0.84
YOLOv2 [3]	76.79	61.31	50.25	0.01
Faster R-CNN with FPN backbone [4]	88.62	84.14	73.20	0.92
MS-CNN [37]	90.03	89.02	76.11	0.18
Improving faster R-CNN [26]	89.20	87.86	74.72	0.06
SINet [25]	89.60	90.60	77.75	0.12
Multitask CNN [20]	91.28	91.67	85.43	—
Proposed method	93.35	92.18	81.32	0.13