Optics and Precision Engineering, Volume. 31, Issue 13, 2000(2023)

Semi-supervised instance object detection method based on SVD co-training

Rui WANG1,*... Siyang FAN1, Jingwen XU1 and Zhiqing WEN2 |Show fewer author(s)
Author Affiliations
  • 1Laboratory of Precision Opto-mechatronics Technology, Ministry of Education, Institute of Instrumentation Science and Opto-electronics Engineering, Beihang University, Beijing009, China
  • 2Engineering Research Center for Intelligent Robotics, Ji Hua Laboratory, Foshan5800, China
  • show less
    Figures & Tables(9)
    Flow chart of training SVD-RCNN
    Framework of key sample selection
    Comparison of Gram matrix before and after convergence, decomposition and finetuning
    Iteration accuracy of SVD-RCNN with different δacc
    Detection accuracy of pre-training SVD-RCNN on GMU dataset
    Detection result on GMU, AVD and BHID datasets
    • Table 1. Results of ablation experiment

      View table
      View in Article

      Table 1. Results of ablation experiment

      数据集关键样本数量收敛-分解-微调自适应的自标注mAP/%
      GMU19972.1
      19977.6
      19975.8
      19979.3
    • Table 2. Comparison of detection accuracy of training methods on different datasets

      View table
      View in Article

      Table 2. Comparison of detection accuracy of training methods on different datasets

      数据集关键样本数量方 法mAP/%
      GMU199半监督SVD-RCNN79.3
      3 851全监督Faster RCNN81.3
      AVD255半监督SVD-RCNN75.6
      5 083全监督Faster RCNN78.5
      BHID55半监督SVD-RCNN76.1
      1 110全监督Faster RCNN78.4
    • Table 3. Comparison of semi-supervised methods on GMU

      View table
      View in Article

      Table 3. Comparison of semi-supervised methods on GMU

      数据集

      (标注样本数量)

      基础网络方 法mAP/%

      GMU

      (199)

      Faster RCNNUnbiased Teacher776.5
      Soft Teacher1978.3
      本文算法79.3
    Tools

    Get Citation

    Copy Citation Text

    Rui WANG, Siyang FAN, Jingwen XU, Zhiqing WEN. Semi-supervised instance object detection method based on SVD co-training[J]. Optics and Precision Engineering, 2023, 31(13): 2000

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Information Sciences

    Received: Jul. 27, 2022

    Accepted: --

    Published Online: Jul. 26, 2023

    The Author Email: WANG Rui (wangr@buaa.edu.cn)

    DOI:10.37188/OPE.20233113.2000

    Topics