Does Computer Vision Models Understand Human Interpretable Concepts? Benchmarking With Counterfactual Images.