Opto-Electronic Engineering, Volume. 51, Issue 10, 240174(2024)

GLCrowd: a weakly supervised global-local attention model for congested crowd counting

Hongmin Zhang*... Qianqian Tian, Dingding Yan and Lingyu Bu |Show fewer author(s)
Author Affiliations
  • School of Electrical and Electronic Engineering,Chongqing University of Technology,Chongqing 400054,China
  • show less
    Figures & Tables(12)
    GLCrowd network structure
    Global-local attention module
    Comparison of different methods. (a) Traditional convolution; (b) Standard self-attention; (c) Local attention
    ConvFFN moudle
    Partial visualization results
    • Table 1. Information of datasets

      View table
      View in Article

      Table 1. Information of datasets

      Shanghai TechUCF-QNRFUCF_CC_50
      Part APart B
      图像数量482716153550
      平均尺寸589×868768×10242013×29022101×2888
      平均人数501123.612791278
      最大人数3139578128654543
      总人数24167788488125164263974
    • Table 2. Information of experimental environment

      View table
      View in Article

      Table 2. Information of experimental environment

      配置参数
      操作系统Ubuntu 20.04.3 LTS (GNU/Linux 5.15.0-97-generic x86_64)
      显卡型号NVIDIA GeForce RTX 4090(×1)
      显存大小24 G
    • Table 3. Data of partial experimental parameters

      View table
      View in Article

      Table 3. Data of partial experimental parameters

      实验参数具体参数
      权重衰减5×10−4
      训练总周期数1200
      优化器动量0.95
      学习率1×10−5
    • Table 4. Experimental results on Shanghai Tech dataset

      View table
      View in Article

      Table 4. Experimental results on Shanghai Tech dataset

      算法模型Part APart B
      MAEMSEMAEMSE
      HLNet[40]71.5108.611.320
      Transcrowd_gap[29]66.1105.19.316.1
      Transcrowd_token[29]69.0116.510.619.7
      MATT[41]80.1129.411.717.5
      OT_M[20]70.7114.58.113.1
      SUA_crowd[42]68.5121.914.120.6
      PDDNet[43]72.6112.210.317.0
      GLCrowd64.887104.4118.95816.202
    • Table 5. Experimental results on UCF_QNRF dataset

      View table
      View in Article

      Table 5. Experimental results on UCF_QNRF dataset

      算法模型MAEMSE
      HLNet[40]100.4182.6
      Transcrowd_token[29]98.9176.1
      Transcrowd_gap[29]97.2168.5
      OT_M[20]100.6167.6
      SUA_crowd[42]130.3226.3
      PDDNet[43]130.2246.6
      GLCrowd95.523173.453
    • Table 6. Experimental results on UCF_CC_50 dataset

      View table
      View in Article

      Table 6. Experimental results on UCF_CC_50 dataset

      算法模型MAEMSE
      MATT[41]355.0550.0
      CCTrans[31]245.0343.0
      SSGP_Crowd[44]355.0505.0
      SE Cycle GAN[45]373.4528.8
      CDPL_crowd[46]336.5486.1
      Transcrowd[29]272.2395.3
      CrowdFormer[47]218.8330.4
      PC-Net[48]217.3309.7
      GLCrowd209.660282.217
    • Table 7. Comparative results of ablation experiments

      View table
      View in Article

      Table 7. Comparative results of ablation experiments

      局部 注意力ConvFFN回归 令牌Part APart BUCF_CC_50
      MAEMSEMAEMSEMAEMSE
      第一组××75.260125.44110.00319.315240.120327.796
      第二组×67.403108.2989.42818.446236.233305.122
      第三组64.887104.4118.95516.202209.660282.217
    Tools

    Get Citation

    Copy Citation Text

    Hongmin Zhang, Qianqian Tian, Dingding Yan, Lingyu Bu. GLCrowd: a weakly supervised global-local attention model for congested crowd counting[J]. Opto-Electronic Engineering, 2024, 51(10): 240174

    Download Citation

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

    Category: Article

    Received: Jul. 24, 2024

    Accepted: Oct. 8, 2024

    Published Online: Jan. 2, 2025

    The Author Email: Zhang Hongmin (张红民)

    DOI:10.12086/oee.2024.240174

    Topics