Laser & Optoelectronics Progress, Volume. 59, Issue 24, 2420001(2022)

Soft Pseudo-Label and Multi-Scale Feature Fusion for Person Re-Identification

Hao Chen1, Baohua Zhang1,3、*, Xiaoqi Lü2,3, Yu Gu1,3, Yueming Wang1,3, Xin Liu1,3, Yan Ren1, Jianjun Li1,3, and Ming Zhang1,3
Author Affiliations
  • 1School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou 014010, Inner Mongolia, China
  • 2School of Information Engineering, Mongolia Industrial University, Huhehaote010051, Inner Mongolia, China
  • 3Inner Mongolia Key Laboratory of Patten Recognition and Intelligent Image Processing, Baotou 014010, Inner Mongolia, China
  • show less
    Figures & Tables(8)
    Model of soft pseudo-label correction based on improved SPGAN
    Generator structure diagram for SPGAN
    Multiple scale feature reconstruction model
    Visual examples of image-image translation
    Visual collation map of model characteristic pattern
    • Table 1. Results of soft pseudo-label correction method verification experiment

      View table

      Table 1. Results of soft pseudo-label correction method verification experiment

      MethodMarket1501→Duke-MTMC-reIDDuke-MTMC-reID→Market1501
      mAPRank 1Rank 5Rank 10mAPRank 1Rank 5Rank 10
      ResNet-50-DBSCAN54.271.182.586.060.875.084.789.4
      Soft pseudo-label-ResNet-50-DBSCAN63.577.587.693.171.488.395.697.5
      Soft pseudo-label-IBN-a-ResNet-50-DBSCAN67.280.289.992.776.689.795.497.7
    • Table 2. Results of pseudo-label clustering algorithm based on the SPGAN model verification experiment

      View table

      Table 2. Results of pseudo-label clustering algorithm based on the SPGAN model verification experiment

      MethodMarket1501→Duke-MTMC-reIDDuke-MTMC-reID→Market1501
      mAPRank 1Rank 5Rank 10mAPRank 1Rank 5Rank 10
      Soft pseudo-label-IBN-a-ResNet-50-DBSCAN67.280.289.992.776.689.795.497.7
      SPGAN-soft pseudo-label-IBN-a-ResNet-50-DBSCAN69.381.690.493.579.290.994.597.6
      Improved SPGAN-Soft pseudo-label-IBN-a-ResNet-50-DBSCAN70.285.892.195.780.492.597.298.4
    • Table 3. Comparison of unsupervised person recognition accuracy with related methods

      View table

      Table 3. Comparison of unsupervised person recognition accuracy with related methods

      MethodMarket1501→Duke-MTMC-reIDDuke-MTMC-reID→Market1501
      mAPRank 1Rank 5Rank 10mAPRank 1Rank 5Rank 10
      PAD-Net1745.163.277.082.547.675.286.390.2
      MMT-700/5002068.781.891.293.476.590.996.497.9
      AE2246.767.979.283.658.081.691.994.6
      Co-teaching-5001861.777.688.090.771.787.895.096.5
      ECN2140.463.375.880.443.075.187.691.6
      AD-Cluster1954.172.682.585.568.386.794.496.5
      PCB-PAST2354.372.4--54.678.4--
      SSG1653.473.080.683.258.380.090.092.4
      Proposed method70.285.892.195.780.492.597.298.4
    Tools

    Get Citation

    Copy Citation Text

    Hao Chen, Baohua Zhang, Xiaoqi Lü, Yu Gu, Yueming Wang, Xin Liu, Yan Ren, Jianjun Li, Ming Zhang. Soft Pseudo-Label and Multi-Scale Feature Fusion for Person Re-Identification[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2420001

    Download Citation

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

    Category: Optics in Computing

    Received: Sep. 17, 2021

    Accepted: Oct. 29, 2021

    Published Online: Nov. 28, 2022

    The Author Email: Zhang Baohua (zbh_wj2004@imust.cn)

    DOI:10.3788/LOP202259.2420001

    Topics