fzuo / aepecker Goto Github PK

Hi

I try to display your samples used in your demo but I think you already normalized these data. If so, can you let me know what mu and std that you use to normalize your dataset? or How to plot/display the samples used for your demonstration?

Thanks

Hi

Does your method work with a large-scale dataset - ImageNet? I do not see the result for ImageNet in your paper.

Thanks

Viet

Hi

Can you share your code implemented for inpainting algorithm? I want to test my attack by using your method. I need to convert it to pytorch since my code works on pytorch instead of keras and tensorflow.

Thanks

Hi @fzuo

I am implementing your algo #1 by following your paper. There is one problem that I have no idea about. You described connected neighborhood N(.) but you did not describe how it looks like. I am not sure what you mean connected neighbor. Do you mean the neighborhood of a pixel is all pixels around that pixel? For example, 8 pixels around 1 pixel. Is it right?
And how large that neighborhood is? I think it can be extended to more pixels around one pixel but not sure in your paper.

Thanks