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

Table 5

Average MAP of two datasets and comparison of their model sizes.

		YOLOv2-Tiny	YOLOv3-Tiny	YOLOv4-Tiny	JRL-YOLO
Dangerous target dataset	Model size (MB)	42.1	33.1	22.4	5.84
	MAP (%)	73.29	81.81	84.49	85.03
Pascal VOC2007 dataset	Model size (MB)	42.2	33.2	22.5	6.58
	MAP (%)	35.43	39.59	46.50	48.88

The best result is shown in bold.