admin管理员组

文章数量:1794759

YoloV5改进策略:独家原创,LSKA(大可分离核注意力)改进YoloV5,比Transformer更有效

摘要

本文给大家带来一种超大核注意力机制的改进方法,尝试了多种改进方法。不仅速度快,而且还有不同程度的提升了精度!

链接如下:

代码语言:javascript代码运行次数:0运行复制
.2014.3001.5502

YoloV5官方代码测试结果

代码语言:javascript代码运行次数:0运行复制
YOLOv5l summary: 267 layers, 46275213 parameters, 0 gradients, 108.2 GFLOPs
                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 15/15 [00:02<00:00,  5.16it/s]
                   all        230       1412      0.971       0.93      0.986      0.729
                   c17        230        131      0.992      0.992      0.995      0.797
                    c5        230         68      0.953          1      0.994       0.81
            helicopter        230         43      0.974      0.907      0.948       0.57
                  c130        230         85          1      0.981      0.994       0.66
                   f16        230         57      0.999       0.93      0.975      0.677
                    b2        230          2      0.971          1      0.995      0.746
                 other        230         86      0.987      0.915      0.974      0.545
                   b52        230         70      0.983      0.957      0.981      0.803
                  kc10        230         62          1      0.977      0.985      0.819
               command        230         40      0.971          1      0.986      0.782
                   f15        230        123      0.992      0.976      0.994      0.655
                 kc135        230         91      0.988      0.989      0.986      0.699
                   a10        230         27          1      0.526      0.912      0.391
                    b1        230         20      0.949          1      0.995      0.719
                   aew        230         25      0.952          1      0.993      0.781
                   f22        230         17      0.901          1      0.995      0.763
                    p3        230        105      0.997       0.99      0.995      0.789
                    p8        230          1      0.885          1      0.995      0.697
                   f35        230         32      0.969      0.984      0.985      0.569
                   f18        230        125      0.974      0.992       0.99      0.806
                   v22        230         41      0.994          1      0.995      0.641
                 su-27        230         31      0.987          1      0.995      0.842
                 il-38        230         27      0.994          1      0.995      0.785
                tu-134        230          1      0.879          1      0.995      0.796
                 su-33        230          2          1          0      0.995      0.846
                 an-70        230          2      0.943          1      0.995      0.895
                 tu-22        230         98      0.983          1      0.995      0.788

改进一

测试结果

代码语言:javascript代码运行次数:0运行复制
YOLOv5l summary: 347 layers, 55583757 parameters, 0 gradients, 129.2 GFLOPs
                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 15/15 [00:03<00:00,  3.97it/s]
                   all        230       1412      0.969      0.948      0.991      0.725
                   c17        230        131      0.986      0.992      0.995       0.81
                    c5        230         68      0.954          1      0.995      0.829
            helicopter        230         43      0.976      0.963       0.99      0.583
                  c130        230         85          1      0.975      0.994      0.656
                   f16        230         57      0.918      0.947      0.981      0.668
                    b2        230          2      0.972          1      0.995      0.895
                 other        230         86      0.981      0.907      0.981      0.538
                   b52        230         70      0.936      0.971      0.984      0.788
                  kc10        230         62          1      0.981      0.985      0.814
               command        230         40      0.993          1      0.995      0.783
                   f15        230        123      0.991          1      0.995      0.687
                 kc135        230         91      0.986      0.978      0.977      0.686
                   a10        230         27          1      0.963      0.991      0.433
                    b1        230         20      0.975          1      0.995      0.716
                   aew        230         25      0.947          1      0.972      0.747
                   f22        230         17      0.981          1      0.995      0.783
                    p3        230        105      0.998      0.981      0.994      0.789
                    p8        230          1      0.869          1      0.995      0.697
                   f35        230         32      0.982      0.969      0.993       0.54
                   f18        230        125      0.984      0.961      0.991       0.82
                   v22        230         41      0.993          1      0.995      0.684
                 su-27        230         31      0.988          1      0.995      0.845
                 il-38        230         27      0.988          1      0.995      0.788
                tu-134        230          1      0.883          1      0.995      0.796
                 su-33        230          2          1          0      0.995      0.697
                 an-70        230          2      0.906          1      0.995      0.721
                 tu-22        230         98      0.973          1      0.995      0.791

mAP50上升,mAP50-95有所下降!

改进二

测试结果

代码语言:javascript代码运行次数:0运行复制
 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 15/15 [00:04<00:00,  3.49it/s]
                   all        230       1412       0.97      0.949       0.99      0.704
                   c17        230        131      0.978      0.998      0.995      0.809
                    c5        230         68      0.974          1      0.995      0.807
            helicopter        230         43      0.949          1      0.977      0.613
                  c130        230         85      0.999          1      0.995      0.665
                   f16        230         57      0.983      0.965      0.985      0.683
                    b2        230          2      0.928          1      0.995      0.497
                 other        230         86      0.941      0.953      0.943      0.475
                   b52        230         70      0.987      0.971      0.985       0.81
                  kc10        230         62      0.996      0.984      0.985      0.825
               command        230         40      0.991          1      0.995      0.781
                   f15        230        123      0.967      0.992      0.994      0.661
                 kc135        230         91       0.99      0.989      0.986      0.671
                   a10        230         27          1      0.814      0.982      0.425
                    b1        230         20      0.984          1      0.995      0.672
                   aew        230         25      0.921          1      0.986      0.748
                   f22        230         17      0.973          1      0.995      0.708
                    p3        230        105      0.991      0.962      0.991      0.778
                    p8        230          1      0.859          1      0.995      0.398
                   f35        230         32      0.966          1      0.993      0.576
                   f18        230        125      0.989      0.992       0.99      0.805
                   v22        230         41      0.994          1      0.995      0.669
                 su-27        230         31      0.987          1      0.995      0.844
                 il-38        230         27      0.984          1      0.995        0.8
                tu-134        230          1       0.96          1      0.995      0.895
                 su-33        230          2          1          0      0.995      0.846
                 an-70        230          2      0.911          1      0.995      0.746
                 tu-22        230         98      0.995          1      0.995      0.797
Results saved to runs\train\exp4

改进三

测试结果

代码语言:javascript代码运行次数:0运行复制
YOLOv5l summary: 415 layers, 24572181 parameters, 0 gradients, 52.0 GFLOPs
                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 15/15 [00:03<00:00,  4.55it/s]
                   all        230       1412      0.964      0.931      0.987      0.721
                   c17        230        131      0.965      0.992      0.995      0.818
                    c5        230         68      0.984      0.956      0.993      0.838
            helicopter        230         43      0.966      0.977      0.984      0.588
                  c130        230         85      0.991          1      0.995      0.661
                   f16        230         57      0.999      0.965      0.988      0.671
                    b2        230          2      0.929          1      0.995      0.721
                 other        230         86      0.988      0.952      0.979      0.525
                   b52        230         70      0.981      0.971      0.985      0.816
                  kc10        230         62          1      0.984      0.986      0.819
               command        230         40      0.993          1      0.995      0.811
                   f15        230        123      0.903      0.981      0.991      0.684
                 kc135        230         91      0.978      0.989      0.986       0.72
                   a10        230         27          1      0.495      0.877      0.379
                    b1        230         20          1      0.982      0.995      0.692
                   aew        230         25      0.948          1      0.995       0.76
                   f22        230         17       0.88          1      0.982      0.764
                    p3        230        105      0.993      0.981      0.994       0.79
                    p8        230          1      0.863          1      0.995      0.697
                   f35        230         32          1      0.928      0.995      0.536
                   f18        230        125      0.989      0.992       0.99      0.813
                   v22        230         41      0.992          1      0.995      0.722
                 su-27        230         31      0.958          1      0.995      0.808
                 il-38        230         27      0.957          1      0.995      0.828
                tu-134        230          1      0.861          1      0.995      0.796
                 su-33        230          2          1          0      0.995      0.697
                 an-70        230          2      0.924          1      0.995      0.721
                 tu-22        230         98      0.988          1      0.995      0.787

改进四

测试结果

代码语言:javascript代码运行次数:0运行复制

     Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 15/15 [00:03<00:00,  4.21it/s]
                   all        230       1412      0.967      0.944      0.978      0.729
                   c17        230        131      0.984      0.992      0.995      0.818
                    c5        230         68      0.957          1      0.993      0.835
            helicopter        230         43      0.967          1      0.981      0.647
                  c130        230         85      0.965          1      0.993      0.658
                   f16        230         57      0.995      0.965       0.99      0.656
                    b2        230          2      0.933          1      0.995      0.821
                 other        230         86      0.964      0.923      0.953      0.521
                   b52        230         70      0.981      0.971      0.976      0.808
                  kc10        230         62      0.981      0.984      0.986      0.816
               command        230         40      0.992          1      0.995      0.816
                   f15        230        123      0.967          1      0.995      0.684
                 kc135        230         91      0.993      0.989      0.986      0.702
                   a10        230         27          1      0.708      0.981      0.482
                    b1        230         20      0.946          1      0.988      0.753
                   aew        230         25       0.95          1      0.992      0.772
                   f22        230         17      0.973          1      0.995      0.744
                    p3        230        105      0.994      0.962       0.99      0.793
                    p8        230          1       0.89          1      0.995      0.697
                   f35        230         32      0.998          1      0.995      0.587
                   f18        230        125      0.982      0.992       0.99      0.815
                   v22        230         41      0.995          1      0.995      0.713
                 su-27        230         31      0.959          1      0.995      0.827
                 il-38        230         27       0.99          1      0.995      0.848
                tu-134        230          1      0.837          1      0.995      0.895
                 su-33        230          2          1          0      0.662       0.43
                 an-70        230          2      0.923          1      0.995      0.746
                 tu-22        230         98      0.996          1      0.995      0.805

改进五

测试结果

代码语言:javascript代码运行次数:0运行复制
  Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 15/15 [00:03<00:00,  4.92it/s]
                   all        230       1412      0.972      0.946      0.991      0.724
                   c17        230        131      0.991      0.992      0.995      0.803
                    c5        230         68      0.948          1      0.995      0.832
            helicopter        230         43      0.955       0.99      0.971      0.578
                  c130        230         85      0.996      0.988      0.995      0.678
                   f16        230         57      0.972      0.965      0.993       0.69
                    b2        230          2      0.984          1      0.995      0.697
                 other        230         86      0.976      0.945      0.971      0.531
                   b52        230         70      0.969      0.971      0.983      0.804
                  kc10        230         62      0.999      0.984      0.985       0.81
               command        230         40      0.994          1      0.995      0.788
                   f15        230        123       0.99          1      0.995      0.671
                 kc135        230         91      0.984      0.978       0.98      0.693
                   a10        230         27          1      0.755       0.99      0.425
                    b1        230         20      0.963          1      0.995      0.674
                   aew        230         25      0.935          1      0.993      0.732
                   f22        230         17      0.985          1      0.995      0.716
                    p3        230        105       0.99      0.984      0.995      0.794
                    p8        230          1      0.894          1      0.995      0.796
                   f35        230         32          1      0.996      0.995      0.552
                   f18        230        125      0.992      0.982      0.988      0.807
                   v22        230         41      0.994          1      0.995      0.682
                 su-27        230         31      0.965          1      0.995      0.833
                 il-38        230         27       0.98          1      0.995      0.806
                tu-134        230          1      0.894          1      0.995      0.895
                 su-33        230          2          1          0      0.995      0.696
                 an-70        230          2      0.916          1      0.995      0.796
                 tu-22        230         98      0.983          1      0.995      0.777
本文参与 腾讯云自媒体同步曝光计划,分享自微信公众号。原始发表:2023-12-07,如有侵权请联系 cloudcommunity@tencent 删除测试

本文标签: YoloV5改进策略独家原创,LSKA(大可分离核注意力)改进YoloV5,比Transformer更有效