Given the wide variety of paintings throughout history, the stylistic differences between painting genres is highly subjective and often requires the judgment of human experts. Using the dataset Best Artworks of All Time and extracting features from paintings using convolutional neural networks (CNN’s), this project measures the effectiveness of more traditional supervised classifiers to identify a genre within Impressionism, Post-Impressionism, Renaissance and Baroque.

Overall, genre could be successfully predicted from extracted features, with an average AUC of 0.95 in one vs. all classification.

In examining some of the failure cases, it was found that there were clearly human-identifiable features in paintings that correspond to the incorrect classification.

For more details, see the poster and paper below.