Revealing Similar Semantics Inside CNNs: An Interpretable Concept-based Comparison of Feature Spaces