Optics and Precision Engineering, Volume. 30, Issue 13, 1620(2022)
Relocation non-maximum suppression algorithm
Figures & Tables(15)
Fig. 7. Partial experimental results of COCO test dataset on YOLOv3
Fig. 11. Relationship between classification confidence and positioning accuracy
Table 1. [in Chinese]
View tableView in ArticleTable 1. [in Chinese]
输入：候选框集合
输出：最优边界框集合
1：候选框集合
2：利用当前最优边界框
3：计算当前最优边界框
4：利用新的最优边界框
5：递归执行步骤1至4直到候选框集合
Table 1. Experimental results of three different NMS algorithms on VOC test dataset
View tableView in ArticleTable 1. Experimental results of three different NMS algorithms on VOC test dataset
Method mAP@0.5：0.95 mAP@0.5 Recall Precision YOLOv3+NMS
YOLOv3+Soft-NMS
YOLOv3+R-NMS
YOLOv5+NMS
YOLOv5+Soft-NMS
YOLOv5+R-NMS
47.0
47.2
47.7
56.7
56.8
57.2
67.2
67.6
67.6
75.0
75.1
75.3
72.5
72.6
72.7
78.1
77.7
78.4
80.6
80.3
80.9
82.6
82.7
83.0
Table 2. Experimental results of three different NMS algorithms on COCO test dataset
View tableView in ArticleTable 2. Experimental results of three different NMS algorithms on COCO test dataset
Method mAP@0.5：0.95 mAP@0.5 Recall Precision/% YOLOv3+NMS
YOLOv3+Soft-NMS
YOLOv3+R-NMS
YOLOv5+NMS
YOLOv5+Soft-NMS
YOLOv5+R-NMS
34.8
34.8
35.2
42.0
42.0
42.5
53.0
53.4
53.5
60.8
61.3
61.4
59.2
59.7
59.9
61.8
62.0
62.1
65.9
65.6
66.4
74.6
73.9
74.8
Table 3. 和在R-NMS算法上的实验对比
View tableView in ArticleTable 3. 和在R-NMS算法上的实验对比
Method mAP@0.5∶0.95 mAP@0.5 Recall Precision YOLOv3+R-NMS（IoU） 47.5 67.3 72.7 80.6 YOLOv3+R-NMS（P）
YOLOv5+R-NMS（IoU）
YOLOv5+R-NMS（P）
47.7
57.1
57.2
67.6
75.1
75.3
72.7
78.3
78.4
80.9
82.9
83.0
Tools
Get Citation
Copy Citation Text
Shuzhi SU, Runbin CHEN, Yanmin ZHU, Bowen JIANG. Relocation non-maximum suppression algorithm[J]. Optics and Precision Engineering, 2022, 30(13): 1620
Paper Information
Category: Information Sciences
Received: Dec. 24, 2021
Accepted: --
Published Online: Jul. 27, 2022
The Author Email: SU Shuzhi (sushuzhi@foxmail.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20