zju-vipa / netgraft Goto Github PK

Thank for your excellent works for few shot pruning, and I noticed that there is a Fig4 in paper("Learning with different numbers of samples"). Is the "samples" means samples of each class? Since I think the code indicates it means that.

deal author
i am interested in the implementation of grafting ResNet on imagenet, which is also part of your experiment. if released, it would help a lot when transfering the method in my project.
wish to get your help.
in case of inconvenience in contact, my email is [email protected].

你好，尊敬的作者，论文中描述的适应层中的两个1x1卷积可以合并到block中，请问合并的原理请问可以描述下嘛，以及我好像没有在你项目中找到合并代码部分

Thank you for your code, it is an impressive work. But I'm confused about the pretrained model, for example, the teacher model is trained on full CIFAR10, while the few-shot dataset is still the CIFAR10. There seems to be an intersection between the two dataset(both cifar10). And this seems to go against the few-shot problem. I mean, in few-shot setting, the teacher model is supposed to be trained on the partial categories of CIFAR10, and the rest categories as few shot dataset to train the student model.