qizhangli / nobox-attacks Goto Github PK

I couldn't understand why we use a random permutation of the first 500 classes rather than just using label_ind = [i for i in range(500)] to generate the selected_data.csv. I didn't find any use-case for doing this randomly. Please explain if there was any reason for doing so

Can you please tell me the version of DenseNet, WRN and PNASNet used for getting the results in Table 1. Furthermore which version of SENet is used? I couldn't find it in torchvision models.

Very insightful work! Could you please share the filenames of randomly selected 5000 benign ImageNet images? I cannot reproduce the Table3 in supplementary material. Thank you so much~