Table 7. Comparison of the performance of different detection algorithms on the VisDrone2019 dataset.

Model	mAP/%	Parameter/million	t/s
YOLOv4	42.79	63.99	0.025
Faster-RCNN (ResNet50)	33.24	70.55	0.062
YOLOv5l	42.10	46.16	0.020
YOLOR	43.17	52.51	0.023
FPN	32.20	—	—
Cascade R-CNN	31.91	—	—
RetinaNet	21.37	—	—
YOLOv4-ours	52.76	44.79	0.028

Note:

The best results are in bold.