Material for the paper "Why do deep convolutional networks generalize so poorly to small image transformations?" by Aharon Azulay and Yair Weiss.

Implemented with PyTorch

CheckTranslationInvariance.py - creates figures with similar data as in table 1 of the paper.

InpaintingFunction.py - takes an image with black background as input and produces an inpainted image (used in the inpainting protocol).