Normal Maps vs. Visible Images: Comparing Classifiers and Combining Modalities