search

BERYLSHEEP / AdvStyle

37 stars 5 forks source link

How to train your network? #2

Open zhanjiahui opened 3 years ago

zhanjiahui commented 3 years ago

Your paper is pretty good !!

  1. In your paper, I note that some components are fixed and some components are trainable. However, I haven't found any code in your project which represents training some components. Could you please tell me how to train it?
  2. I know that thelta is an important trainable parameter, but I am fused about if alpha is trainable?
Johnson-yue commented 2 years ago

+1

zhouxiaowei1120 commented 1 year ago

+1

bigorange-1 commented 1 year ago

Hello, for your Resnet, is it training in a way of Siamese network? Because , i found that you calculate the similarity score of two images in your paper. That seems like Siamese network. I can't found any training code. Can you tell me how to train? Thanks

bigorange-1 commented 1 year ago

Maybe i am wrong. I can't understand the usage of Resnet18. In 3.2.2, you used the cross-entropy. In 3.3.2，you calculated the identity loss by the cosine distance function. Firstly, training the network in classifier. Then, Extracting the features of images using the classifier. Is this right? Could you tell me which one is right? Thank you very much.