caffe（五）——windows下输出log日志和画出accuracy和loss曲线

这里注意运行的python版本，我下的caffe对应的是python2.7

参考：https://www.zhihu.com/question/49521165/answer/127675889

1.生成日志

caffe.exe train --solver=solver.prototxt --weights=model.caffemodel >train.log 2>&1

2.画图

(1)将 caffe-master\tools\extra\ 文件夹下的 parse_log.py、extract_seconds.py 、plot_training_log.py 还有 train.log 放到同一文件夹下

(2)修改 “plot_training_log.py” 文件

• get_log_parsing_script()

• create_field_index()

field_index = {train_key:{'Iters':0, 'Seconds':1, train_key + ' learning rate':2,
                              train_key + ' loss':3},
                   test_key:{'Iters':0, 'Seconds':1, test_key + ' learning rate':2,
                   test_key + ' accuracy':3, test_key + ' loss':4}}

• load_data()

(3)python parse_log.py train.log ./ 在当前目录下生成 .train、.test 文件

注意：之前很奇怪，不生成.train .test文件，后来发现是solver.prototxt中 display 参数的问题，应该是通过这个隔几次记录 accuracy、loss 值

(4)画图

python plot_training_log.py 0 accuracy.png train.log

python plot_training_log.py ６ lr.png train.log

python plot_training_log.py 8 loss.png train.log

数字参数对应的输出图的类型与plot_training_log.py中修改的create_field_index()有关

Supported chart types
0: Test accuracy vs. Iters
1: Test accuracy vs. Seconds
2: Test learning rate vs. Iters
3: Test learning rate vs. Seconds
4: Test loss vs. Iters
5: Test loss vs. Seconds
6: Train learning rate vs. Iters
7: Train learning rate vs. Seconds
8: Train loss vs. Iters
9: Train loss vs. Seconds