Optics and Precision Engineering, Volume. 32, Issue 20, 3099(2024)
A self correcting low-light object detection method based on pyramid edge enhancement
Figures & Tables(11)
Fig. 8. Comparison of visualization results of different algorithms
Table 1. Comparison of ablation experiments for different categories of AP indicators
View tableView in ArticleTable 1. Comparison of ablation experiments for different categories of AP indicators
Category AP/% RTDETR RTDETR+M1 RTDETR+M2 GL-RTDETR Bicycle 85.01 82.30 80.34 86.12 Boat 61.40 75.42 67.43 75.66 Bottle 59.62 60.37 61.14 59.79 Bus 77.48 79.98 78.48 80.05 Car 73.81 74.47 73.54 74.71 Cat 69.10 67.32 66.21 70.13 Chair 45.51 48.51 47.24 46.34 Cup 74.07 70.43 71.16 73.16 Dog 56.13 61.26 59.06 59.02 Motorbike 66.01 69.63 67.53 70.34 People 76.00 79.57 79.32 77.81 Table 46.52 47.09 46.95 47.42 mAP 65.89 68.03 66.53 68.38
Table 2. Comparison of ablation experiments with different improvement strategies
View tableView in ArticleTable 2. Comparison of ablation experiments with different improvement strategies
RTDETR M1 M2 mAP/% mmAP/% R/% Para/M MS/MB FPS √ 65.89 39.63 56.91 42.80 86.10 86.96 √ √ 68.03 41.51 58.64 42.80 86.10 86.96 √ √ 66.53 40.16 57.15 37.85 72.79 72.23 √ √ √ 68.38 41.97 59.00 37.85 72.79 72.23
Table 3. Experimental comparison results of different methods
View tableView in ArticleTable 3. Experimental comparison results of different methods
GENet Methods mAP/% mmAP/% R/% Para/M MS/MB FPS frame/s × YOLOv5l 64.22 44.14 57.96 46.20 89.21 76.21 YOLOv7x 66.51 43.37 58.16 70.89 136.47 65.79 YOLOv8l 67.61 42.52 58.39 43.64 83.70 78.76 RTDETR 65.89 39.63 56.91 42.80 86.10 86.96 √ YOLOv5l 65.49 44.68 58.73 46.20 89.21 76.21 YOLOv7x 67.43 43.91 58.94 70.89 136.47 65.79 YOLOv8l 68.07 43.65 58.97 43.64 83.70 78.76 GL-RTDETR 68.38 41.97 59.00 37.85 72.79 72.23 / ZERO-DCE 68.04 43.84 58.74 44.50 147.36 66.29 / IA-YOLO 67.89 43.49 58.45 62.31 156.18 49.46 / IAT 66.54 42.17 58.41 57.49 173.71 42.49
Tools
Get Citation
Copy Citation Text
Zhanjun JIANG, Baijing WU, Long MA, Jing LIAN. A self correcting low-light object detection method based on pyramid edge enhancement[J]. Optics and Precision Engineering, 2024, 32(20): 3099
Paper Information
Category:
Received: Apr. 30, 2024
Accepted: --
Published Online: Jan. 10, 2025
The Author Email: WU Baijing (12211816@ stu.lzjtu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20