• Laser & Optoelectronics Progress
  • Vol. 58, Issue 12, 1210013 (2021)
Junkai Luo, Baohua Zhang*, Yanyue Zhang, Yu Gu, Yueming Wang, Xin Liu, Yan Ren, Jianjun Li, and Ming Zhang
Author Affiliations
  • College of Information Engineering, Inner Mongolia University of Science & Technology, Baotou, Inner Mongolia 014010, China
  • show less
    DOI: 10.3788/LOP202158.1210013 Cite this Article Set citation alerts
    Junkai Luo, Baohua Zhang, Yanyue Zhang, Yu Gu, Yueming Wang, Xin Liu, Yan Ren, Jianjun Li, Ming Zhang. An Object Detection Algorithm Based on Contextual Self-Calibration And Dual-Attention Mechanism[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210013 Copy Citation Text show less
    Contextual self-calibration dual-attention module
    Fig. 1. Contextual self-calibration dual-attention module
    Contextual self-calibrated spatial attention
    Fig. 2. Contextual self-calibrated spatial attention
    Multi-kernel channel attention
    Fig. 3. Multi-kernel channel attention
    Network structure of proposed algorithm
    Fig. 4. Network structure of proposed algorithm
    Multi-class object detection results
    Fig. 5. Multi-class object detection results
    MethodBackbonePAP /%PAP50 /%PAP75 /%PAPs /%PAPm /%PAPl /%
    YOLOv2[3]DarkNet-1921.644.019.25.022.435.5
    YOLOv3[4]DarkNet-5330.057.944.124.144.251.2
    Faster R-CNN[1]ResNet-10136.259.139.018.239.048.2
    SSD513[2]ResNet-10131.250.433.310.234.549.8
    CornerNet[5]Hourglass-10440.563.846.319.442.753.9
    ExtremeNet[6]Hourglass-10443.760.547.024.146.957.6
    Proposed algorithmResNet-10145.264.348.928.349.558.9
    Table 1. Test results of various algorithms on MSCOCO dataset
    CSDAMDCNV2PAP /%Testing time /ms
    42.357
    43.061
    45.073
    45.278
    Table 2. Ablation experiment results
    Junkai Luo, Baohua Zhang, Yanyue Zhang, Yu Gu, Yueming Wang, Xin Liu, Yan Ren, Jianjun Li, Ming Zhang. An Object Detection Algorithm Based on Contextual Self-Calibration And Dual-Attention Mechanism[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210013
    Download Citation