(Question from Oxford Univ) Why do you divide the distance between the new image and the original image by the norm of the original image?

kupl / adapt

ADAPT is the open source white-box testing framework for deep neural networks

MIT License

21 stars 5 forks source link

(Question from Oxford Univ) Why do you divide the distance between the new image and the original image by the norm of the original image? #4

Closed sooyoungcha closed 3 years ago

sooyoungcha commented 3 years ago

"adapt/fuzzer/fuzzer.py line 173" Why do you divide the distance between the new image and the original image by the norm of the original image? I'm not sure what this corresponds to.

sooyoungcha commented 3 years ago

Since the size of the inputs and the valid range of each pixel are different in MNIST and ImageNet, the distances between the original input and the perturbed input in each domain are very different. Therefore, distances are normalized by dividing by the original input.