Laser & Optoelectronics Progress, Volume. 57, Issue 2, 21501(2020)
Rapid Detection Algorithm for Small Objects Based on Receptive Field Block
Figures & Tables(7)
Fig. 4. Comparison of detection results of small objects. (a) Detection results of SSD algorithm; (b) detection results of proposed algorithm
Table 1. Comparison of detection results on PASCAL VOC 2007 test set for different object detection algorithms
View tableTable 1. Comparison of detection results on PASCAL VOC 2007 test set for different object detection algorithms
Method Backbone Dataset mAP /% Frame rate / (frame·s-1) R-FCN[ 16 ]ResNet-101 VOC07+12 80.5 9 Fast YOLO[ 4 ]Google Net-9 VOC07+12 52.7 163 YOLO[ 4 ]VGG-16 VOC07+12 66.4 96 YOLO- V2[ 17 ]DarkNet-19 VOC07+12 78.8 80 DSSD321[ 9 ]ResNet-101 VOC07+12 78.6 9.5 SSD300[ 5 ]VGG-16 VOC07+12 77.2 120 Proposed (300) VGG-16 VOC07+12 80.5 75
Table 2. Influences of different modules on accuracy of algorithms
View tableTable 2. Influences of different modules on accuracy of algorithms
Algorithm SSD SSD+ RFB SSD+ Fusion Proposed model Add RFB? × √ × √ Add Fusion? × × √ √ mAP /% 77.2 80.1 78.9 80.5 FPS 120 83 104 75
Table 3. Detection results of different fusion ways
View tableTable 3. Detection results of different fusion ways
Layer mAP /% Conv4_3 80.1 Conv4_3+Conv5_3 80.5 Conv4_3+Conv6 80.2
Tools
Get Citation
Copy Citation Text
Wang Weifeng, Jin Jie, Chen Jingming. Rapid Detection Algorithm for Small Objects Based on Receptive Field Block[J]. Laser & Optoelectronics Progress, 2020, 57(2): 21501
Paper Information
Category: Machine Vision
Received: Apr. 29, 2019
Accepted: --
Published Online: Jan. 3, 2020
The Author Email: Weifeng Wang (weifeng_17@tju.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20