Table 5.

Comparison of metrics for different detection algorithms on the VisDrone2019 dataset.

Method	Input Image Size	mAP50val (%)	mAP50test (%)	Params (M)	Size (M)	Timeval (ms)	Timetest (ms)	Average Speed/FPS
YOLOv3	640	39.2	32.6	61.55	123.6	25.2	19.2	22.2
	1024	51.9	43.1			38.5	32.2	35.35
YOLOv3-tiny	640	16.5	14.1	8.69	17.5	7.0	5.3	6.15
	1024	29.3	23.6			9.7	7.2	8.45
YOLOX-s	640	32.6	27.7	9.0	8.94	25.6	22.9	24.25
	1024	40.3	34.1			28.7	26.4	27.55
YOLOv7-tiny	640	35.4	30.3	6.03	12.3	24.8	23.6	24.2
	1024	45.2	38.1			26.8	25.4	26.1
YOLOv8n	640	33.4	27.2	3.01	6.2	12.9	6.2	9.55
	1024	43.4	35.9			14.1	7.4	10.75
YOLOv5s	640	33.1	27.8	7.10	14.5	13.2	11.8	12.5
	1024	45.1	38.0			17.8	16.0	16.9
YOLOv5x	640	40.8	34.2	86.28	173.2	28.1	27.0	27.55
	1024	53.2	43.9			64	62.9	63.45
MFP-YOLO	640	46.0	35.8	1.48	3.81	30.8	26.2	28.5
	1024	53.7	44.1			58	52.5	55.25