Feature Importance Methods for Computer Vision Please see final branch for the repo. We explored and compared three methods of computing feature importance: Integrated Gradients Local Data Matrix LIME