About training set - Githubissues

NKUhealong / CABnet

MIT License

39 stars 13 forks source link

About training set #1

Open xia0min opened 3 years ago

xia0min commented 3 years ago

hi, I am so appreciated with your great job. And i have read your code carefully. When i read the train.py, it confused me in " 'DDR': [9851,2503,5,'./data/DDR/train','./data/DDR/valid']"

In your paper,the number of training is 6260,but 9851 in the code. so how do you get 9851 pictures in the training dataset look forward to your reply. thanks.

NKUhealong commented 3 years ago

We augment the classes with fewer training samples, such as label 1,label 3 and label 4, you can augment these categories

xia0min commented 3 years ago

thank you ,i will try it

zhangzhuoneng commented 3 years ago

您好，我仔细看了下你的代码，有点疑惑。为什么您的epochs有两个，学习率也有两个呢。那个epoch=1的作用是啥呢？谢谢！

houjunlin commented 1 year ago

您好，我仔细看了下你的代码，有点疑惑。为什么您的epochs有两个，学习率也有两个呢。那个epoch=1的作用是啥呢？谢谢！

I have the same question. Why set layer.trainable=False and train one epoch?