A problem when training model on ImageNetMini

[czb2133]

Thanks for releasing the code.
I found that you are using cls_id to choose the sample-wise noise from perturb_tensor. Should I change cls_id to the data index of the dataloader to choose the sample-wise noise from perturb_tensor?

[HanxunH]

This may depend on how the noise is generated. I'm not sure what do you mean by using cls_id to choose the sample-wise noise from perturb_tensor.

In ImageNetMini class, I used the cls_id to select the first 100 classes, and use this dataset to generate the noise (perturb_tensor). So the perturb_tensor only contains the noises for images of the first 100 classes. When the noise is applied to the data, the perturbed dataset inherits from ImageNetMini class, the ordering should be the same with the perturb_tensor.