【Visdrone数据集】Vsidrone+Yolov5结果记录

YOLOV5

yolov5s epoch=100

训练

训练代码：yolov5s 100epoch

• 命令

python train.py --data VisDrone.yaml --weights yolov5s.pt --img 640

分辨率：img=640

输出权重文件：exp4

验证集

• 运行验证集命令：

python val.py --weights ./runs/train/exp4/weights/best.pt --data VisDrone.yaml --img 640 --task val

• 验证集结果

       Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 18/18 [00:16<00:00,  1.11it/s]                                               
          all        548      38759      0.443       0.34       0.33       0.18                                                                                                
   pedestrian        548       8844      0.472      0.379      0.388      0.163                                                                                                
       people        548       5125      0.442      0.336      0.322      0.111                                                                                                
      bicycle        548       1287      0.251      0.156      0.114     0.0415                                                                                                
          car        548      14064       0.63      0.719       0.73      0.488                                                                                                
          van        548       1975      0.467      0.354      0.354       0.24                                                                                                
        truck        548        750      0.478      0.301      0.314      0.194                                                                                                
     tricycle        548       1045      0.432      0.197      0.198      0.101                                                                                                
  

  awning-tricycle 548 532 0.26 0.126 0.111 0.0697
  bus 548 251 0.533 0.406 0.388 0.238
  motor 548 4886 0.466 0.426 0.379 0.149

测试集

• 运行测试集命令：

python val.py --weights ./runs/train/exp4/weights/best.pt --data VisDrone.yaml --img 640 --task val

• 测试集结果

             Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 51/51 [00:26<00:00,  1.92it/s]                                               
               all       1610      75102      0.394      0.309      0.282       0.15                                                                                                
        pedestrian       1610      21006      0.415      0.252      0.245     0.0928                                                                                                
            people       1610       6376      0.393      0.172      0.162     0.0511                                                                                                
           bicycle       1610       1302      0.231     0.0891     0.0767     0.0277                                                                                                
               car       1610      28074      0.587      0.715      0.693      0.412                                                                                                
               van       1610       5771       0.35      0.374      0.311       0.19                                                                                                
             truck       1610       2659      0.368      0.389      0.317      0.176                                                                                                
          tricycle       1610        530      0.229      0.175      0.115     0.0561                                                                                                
   awning-tricycle       1610        599      0.364      0.139      0.152     0.0834                                                                                                
               bus       1610       2940       0.62      0.505      0.529      0.337                                                                                                
             motor       1610       5845      0.379      0.282       0.22     0.0781     

yolov5s epoch=300

训练

训练代码：yolov5s 300epoch （直接在train.py中改了epoch默认值）

• 命令

python train.py --data VisDrone.yaml --weights yolov5s.pt --img 640

分辨率：img=640

输出权重文件：exp5

验证集

• 运行验证集命令

 python val.py --weights ./runs/train/exp5/weights/best.pt --data VisDrone.yaml --img 640 --task val

• 验证集结果

             Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 18/18 [00:16<00:00,  1.12it/s]
               all        548      38759      0.466      0.353       0.35      0.194
        pedestrian        548       8844       0.48      0.389      0.401       0.17
            people        548       5125      0.453      0.348      0.332       0.12
           bicycle        548       1287      0.303      0.145      0.129     0.0465
               car        548      14064      0.641      0.727       0.74        0.5
               van        548       1975      0.464      0.376      0.374      0.255
             truck        548        750      0.515      0.317      0.332      0.205
          tricycle        548       1045      0.421      0.224      0.214      0.112
   awning-tricycle        548        532      0.316      0.145      0.125     0.0795
               bus        548        251      0.597      0.426      0.457      0.292
             motor        548       4886      0.474      0.437      0.394       0.16

测试集

• 运行测试集命令

 python val.py --weights ./runs/train/exp5/weights/best.pt --data VisDrone.yaml --img 640 --task test

-测试集结果

                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 51/51 [00:26<00:00,  1.93it/s]
                   all       1610      75102      0.421      0.314      0.294      0.158
            pedestrian       1610      21006       0.43      0.256      0.249      0.095
                people       1610       6376      0.424      0.168      0.166      0.055
               bicycle       1610       1302      0.255      0.073     0.0811     0.0284
                   car       1610      28074      0.605      0.718      0.699      0.419
                   van       1610       5771      0.371      0.365       0.32      0.199
                 truck       1610       2659        0.4      0.384      0.332      0.189
              tricycle       1610        530      0.254      0.209      0.143     0.0664
       awning-tricycle       1610        599      0.415      0.174      0.169     0.0897
                   bus       1610       2940      0.657      0.506      0.546      0.351
                 motor       1610       5845      0.404      0.285      0.233     0.0849