yolov5 【v4.0】用自有数据集训练结果大比拼【5s,5m,5l,5x,5shpy】

2023-10-29

5s:

python test.py --weights runs/train/exp9/weights/best.pt --data data/9.yaml --task test --save-json --iou-thres 0.5
模型大小：best.pt 13.76MB
Class Images Targets P R mAP@.5 mAP@.5:.95
all 144 268 0.912 0.983 0.983 0.899
b6925303773908 144 27 0.968 1 0.995 0.903
b6922255451427 144 32 0.868 0.969 0.963 0.858
b6901285991219 144 27 0.784 0.963 0.948 0.827
b6920459905012 144 40 1 0.979 0.996 0.854
b6921168509256 144 37 0.891 1 0.995 0.902
b6920005772716 144 19 0.954 1 0.995 0.899
b6924882485103 144 17 0.95 1 0.995 0.989
b6920152471616 144 23 0.963 1 0.995 0.968
b6924743915848 144 46 0.832 0.935 0.959 0.89

5shpy:（5s超参数优化貌似没啥提升）

python test.py --weights runs/train/exp18/weights/best.pt --data data/9.yaml --task test --save-json --iou-thres 0.5
模型大小：best.pt 13.76MB

Class Images Targets P R mAP@.5 mAP@.5:.95
all 144 268 0.869 0.989 0.982 0.896
b6925303773908 144 27 0.966 1 0.998 0.917
b6922255451427 144 32 0.804 0.969 0.955 0.865
b6901285991219 144 27 0.63 0.963 0.943 0.82
b6920459905012 144 40 0.975 0.994 0.996 0.834
b6921168509256 144 37 0.848 1 0.996 0.884
b6920005772716 144 19 0.921 1 0.997 0.898
b6924882485103 144 17 0.95 1 0.997 0.984
b6920152471616 144 23 0.963 1 0.997 0.966
b6924743915848 144 46 0.766 0.978 0.961 0.894

5m:

python test.py --weights runs/train/exp15/weights/best.pt --data data/9.yaml --task test --save-json --iou-thres 0.5
   模型大小：best.pt 40.56MB
                  Class Images Targets P R mAP@.5 mAP@.5:.95:
                        all 144 268 0.908 0.99 0.984 0.91
b6925303773908 144 27 0.968 1 0.995 0.924
b6922255451427 144 32 0.864 0.969 0.97 0.906
b6901285991219 144 27 0.803 0.963 0.949 0.848
b6920459905012 144 40 0.952 1 0.996 0.85
b6921168509256 144 37 0.889 1 0.995 0.902
b6920005772716 144 19 0.955 1 0.995 0.919
b6924882485103 144 17 0.95 1 0.995 0.972
b6920152471616 144 23 0.963 1 0.995 0.96
b6924743915848 144 46 0.83 0.978 0.963 0.906

5mhpy:（5m超参数优化提升也不明显，可能是数据集比较小）

python test.py --weights runs/train/exp18/weights/best.pt --data data/9.yaml --task test --save-json --iou-thres 0.5
模型大小：best.pt 40.56MB

Class Images Targets P R mAP@.5 mAP@.5:.95:

all 144 268 0.908 0.99 0.983 0.913
b6925303773908 144 27 0.968 1 0.996 0.932
b6922255451427 144 32 0.869 0.969 0.963 0.886
b6901285991219 144 27 0.756 0.963 0.952 0.831
b6920459905012 144 40 0.983 1 0.996 0.88
b6921168509256 144 37 0.909 1 0.996 0.918
b6920005772716 144 19 0.955 1 0.995 0.931
b6924882485103 144 17 0.95 1 0.995 0.961
b6920152471616 144 23 0.961 1 0.995 0.978
b6924743915848 144 46 0.822 0.978 0.961 0.9

5l:

python test.py --weights runs/train/exp16/weights/best.pt --data data/9.yaml --task test --save-json --iou-thres 0.5
模型大小：best.pt 89.47MB
Class Images Targets P R mAP@.5 mAP@.5:.95:
all 144 268 0.907 0.988 0.983 0.913
b6925303773908 144 27 0.968 1 0.995 0.923
b6922255451427 144 32 0.859 0.955 0.964 0.892
b6901285991219 144 27 0.754 0.963 0.951 0.853
b6920459905012 144 40 1 0.997 0.996 0.868
b6921168509256 144 37 0.882 1 0.995 0.918
b6920005772716 144 19 0.955 1 0.995 0.903
b6924882485103 144 17 0.929 1 0.995 0.978
b6920152471616 144 23 0.962 1 0.995 0.988
b6924743915848 144 46 0.851 0.978 0.958 0.89

5x:

python test.py --weights runs/train/exp17/weights/best.pt --data data/9.yaml --task test --save-json --iou-thres 0.5

模型大小：best.pt 175.2MB

Class Images Targets P R mAP@.5 mAP@.5:.95:
all 144 268 0.921 0.987 0.98 0.916
b6925303773908 144 27 0.968 1 0.995 0.925
b6922255451427 144 32 0.87 0.969 0.967 0.886
b6901285991219 144 27 0.846 0.963 0.941 0.84
b6920459905012 144 40 0.981 0.975 0.976 0.855
b6921168509256 144 37 0.904 1 0.996 0.936
b6920005772716 144 19 0.957 1 0.995 0.947
b6924882485103 144 17 0.95 1 0.995 0.973
b6920152471616 144 23 0.971 1 0.995 0.981
b6924743915848 144 46 0.844 0.978 0.957 0.9

结论：可能是数据集的问题，数据差距不是很大，暂时用5s了

本文内容由网友自发贡献，版权归原作者所有，本站不承担相应法律责任。如您发现有涉嫌抄袭侵权的内容，请联系:hwhale#tublm.com(使用前将#替换为@)