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

Table 1

The YOLOv3-Tiny network structure.

Layer	Type	Filters	Size/stride	Output
0	Convolutional	16	3 × 3/1	416 × 416 × 16
1	Maxpool	—	2 × 2/2	208 × 208 × 16
2	Convolutional	32	3 × 3/1	208 × 208 × 32
3	Maxpool	—	2 × 2/2	104 × 104 × 32
4	Convolutional	64	3 × 3/1	104 × 104 × 64
5	Maxpool	—	2 × 2/2	52 × 52 × 64
6	Convolutional	128	3 × 3/1	52 × 52 × 128
7	Maxpool	—	2 × 2/2	26 × 26 × 128
8	Convolutional	256	3 × 3/1	26 × 26 × 256
9	Maxpool	—	2 × 2/2	13 × 13 × 256
10	Convolutional	512	3 × 3/1	13 × 13 × 512
11	Maxpool	—	2 × 2/1	13 × 13 × 512
12	Convolutional	1024	3 × 3/1	13 × 13 × 1024
13	Convolutional	256	1 × 1/1	13 × 13 × 256
14	Convolutional	512	3 × 3/1	13 × 13 × 512
15	Convolutional	36	1 × 1/1	13 × 13 × 36
16	YOLO	—	—	—
17	Route 13	—	—	—
18	Convolutional	128	1 × 1/1	13 × 13 × 128
19	Upsample	—	2 × 2/1	26 × 26 × 128
20	Route 19, 8	—	—	—
21	Convolutional	256	3 × 3/1	26 × 26 × 256
22	Convolutional	36	1 × 1/1	26 × 26 × 256
23	YOLO	—	—	—