Skip to content

分类任务

Mr.Li edited this page Jun 20, 2022 · 3 revisions

一. 安装依赖

pip install -r ./Package/requirements.txt 

二. 准备数据

根目录已包含CatDog数据集,且Config/已生成dataset.txt

txt格式为[类型, 类别名, 图像路径]

train, dog,  /xxxx/img1.jpg
val,   cat,  /xxxx/img2.jpg
test,  cat,  /xxxx/img3.jpg

三. 训练

  1. Config/config.py训练参数已配好

  2. 开始训练

    假设单机2卡,执行如下命令 访问更多详情

    torchrun --nproc_per_node 2 train.py

    控制台输出 TensorBoard和Checkpoint将保存在/ExpLog/文件夹下

    ****************************
    The nums of trainSet: 140
    The nums of each class:  {'dog': 70, 'cat': 70}
    
    ****************************
    The nums of valSet: 20
    The nums of each class:  {'dog': 10, 'cat': 10} 
    
    start epoch 0/50...
  3. TensorBoard可视化

四. 测试

python test.py     --txt ./Config/dataset.txt  --weights 权重路径

控制台输出 120+指标

****************************
The nums of testSet: 40
The nums of each class:  {'dog': 20, 'cat': 20} 

Overall ACC is 0.900 

matrix save in  /../matrix.jpg 

Predict    cat        dog        
Actual
cat        0.95       0.05       

dog        0.15       0.85       


Predict   cat       dog       
Actual
cat       19        1         

dog       3         17        





Overall Statistics : 

95% CI                                                            (0.80703,0.99297)
ACC Macro                                                         0.9
ARI                                                               0.63062
AUNP                                                              0.9
AUNU                                                              0.9
Bangdiwala B                                                      0.8125
Bennett S                                                         0.8
CBA                                                               0.85682
CSI                                                               0.80404
Chi-Squared                                                       25.85859
Chi-Squared DF                                                    1
Conditional Entropy                                               0.44812
Cramer V                                                          0.80403
Cross Entropy                                                     1.00725
F1 Macro                                                          0.89975
F1 Micro                                                          0.9
...

五. 预测

 python predict.py -weights 权重路径  --vis_cam