Implementation of Spatially Transformed Adversarial Attack

The task is to create adversarial examples using white box attack based threat model.

In this paper, authors propose very unique way of making these adversarial examples, instead of manipulating the pixel values, they apply spatial transformation to the image which are perpetually realistic and difficult to defend. 

I have implemented the first PyToch implementation version of this adversarial attack, and refer tensorflow version implementation from this repo  For getting more info on this implementation visit on my github link. This code have been made public to be used in any research or project freely.