Multi-Scale Feature Fusion Lightweight Real-Time Infrared Pedestrian Detection at Night

Table 1. YOLO-MN model test results
View table
Table 1. YOLO-MN model test results
Network structure Accuracy /% Speed /(frame·s^－1) Model size /MB
YOLOv3 90.2 26 246
YOLO-MN 89.0 62 95

Table 2. YOLO-MN-SPP model test results
View table
Table 2. YOLO-MN-SPP model test results
Network structure Accuracy /% Speed /(frame·s^－1) Model size /MB
YOLO-MN 89.0 62 95.0
YOLO-MN-SPP 90.1 47 99.5
YOLO-MN-SPP1 92.3 39 100.1

Table 3. Pruning results with different pruning rates
View table
Table 3. Pruning results with different pruning rates
Pruning rate /% Accuracy /% Model size /MB Number of channels
85 92.3 13.6 1450
90 92.3 12.4 967
92 92.2 12.1 774
94 92.2 11.8 580
95 91.2 11.7 485
96 29.5 11.6 387

Table 4. YOLO-Person model test results
View table
Table 4. YOLO-Person model test results
Network structure Accuracy /% Speed /(frame·s^－1) Model size /MB
YOLOv3-tiny 71.3 83 33.1
YOLO-Person1 91.2 69 11.7
YOLO-Person 92.2 69 11.7

Table 5. Jetson Nano platform test results
View table
Table 5. Jetson Nano platform test results
Network structure Accuracy /% Speed /(frame·s^－1) Model size /MB
YOLOv3 90.2 1.2 246.0
YOLOv3-tiny 71.3 14.0 33.1
YOLO-Person 92.2 12.0 11.7

Tools

Get Citation

Copy Citation Text

Zifen He, Guangchen Chen, Junsong Chen, Yinhui Zhang. Multi-Scale Feature Fusion Lightweight Real-Time Infrared Pedestrian Detection at Night[J]. Chinese Journal of Lasers, 2022, 49(17): 1709002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites