JRL-YOLO: A Novel Jump-Join Repetitious Learning Structure for Real-Time Dangerous Object Detection

Table 3

Comparison of using three different lightweight models on the dangerous target dataset.

Precision	Classes	YOLOv2-Tiny	YOLOv3-Tiny	JRL-YOLO
AP (%)	Gun	75.58	85.92	89.26
	Knife	38.76	71.66	80.50
	Hit	70.71	81.64	81.77
	Kick	91.68	88.36	94.85
	Fall	85.01	87.20	88.08
	Stick	74.03	81.58	88.18
	Short knife	75.14	77.66	81.25
MAP (%)	—	73.29	81.81	85.03