Why you set "target_sigma = args.sigma * 2" in certify.py(line 20)?

[ChoiDae1]

Hello! Can you explain why you set target_sigma = args.sigma*2 (multiply 2)?
In my opinion, target_sigma = args.sigma**2 (power 2) is correct.

Thank you!

[Eric-mingjie]

Hi, Thanks for your interest in our work. We use args.sigma*2 because the image we loaded is in range of [-1,1]. Thus the corresponding sigma (for image with pixel range [0,1]) needs to be multiplied by 2 to match the normalized images.

[ChoiDae1]

Thank you for answering!
Then, can you explain why you use args.sigma(not args.sigma*2) when computing certified radius?

[Eric-mingjie]

In the original randomized smoothing paper, the certified radius proposed in equation 3 use just sigma. I don't know where you get the impression that sigma**2 is needed, can you point me to a reference?

[ChoiDae1]

I was confused for a second. Thank you!