Python implementation for evaluating explanations presented in On the (In)fidelity and Sensitivity for Explanations published in NeurIPS 2019 for evaluating any saliency explanations.
Run vis_mnist.ipynb to see examples of explanations in MNIST along with their sensitivity and infidelity.
We build our visualization tools based on codes available in the following repositories: