An Efficient Multiscale Pyramid Attention Network for Face Detection in Surveillance Images

Table 5

The runtime and parameters compare our model with different approaches on Wider Face dataset.

Method	Backbone	mAP	Speed	Parameters	Runtime
FasterRCNN [33]	VGG19	71.2	<6 FPS	24M	180 ms
ScaleFace [28]	ResNet50	77.2	<8 FPS	37M	130 ms
HR [19]	ResNet101	80.6	3.1 FPS	58.16 M	360 ms
Gao [31]	TinyYOLOv3	91.5	<6 FPS	33M	182 ms
Ours	Efficient-B3	89.2	24 FPS	15M	80 ms