Multiscale Feature Learning Based on Enhanced Feature Pyramid for Vehicle Detection

Table 4

The AP results and the inference time of each enhanced module and the original Faster R–CNN with FPN backbone.

Model	Average precision (%)			Time/image (s)
	Easy	Moderate	Hard
Faster R–CNN with FPN backbone	89.52	87.14	78.20	0.17
+Improving RPN	89.21	88.27	78.12	0.09
+Improving RPN + multilayer enhancement module	92.71	91.64	80.22	0.11
+Improving RPN + multilayer enhancement module + adaptive RoI pooling	93.67	92.08	82.41	0.13

The inference time is evaluated on single Nvidia RTX 3070 GPU.