Improved YOLOv4 for Pedestrian Detection and Counting in UAV Images

Table 1

The main difference of YOLO series.

Model	Backbone	Feature map	Feature fusion	Number of anchors
YOLO	Conv × 24	7 × 7	-	0
YOLOv2	Darknet-19	13 × 13	-	5
YOLOv3	Darknet-53	13 × 13/26 × 26/52 × 52	FPN	9
YOLOv4	CSPDarknet-53	13 × 13/26 × 26/52 × 52	PANet	9