Learning the Cultural Consistent Facial Aesthetics by Convolutional Neural Network

Studying facial aesthetics has stimulated great interests in psychology and computer science due to a constant debate on whether it is cross-culture coherent or culture specific. Most computational models follow the cross-culture coherence theory and quantify the facial aesthetics by handcrafted geometry and appearance features, however, which are not directly derived from the raw data. In this work, we develop an end-to-end Convolutional Neural Network (CNN) model to recognize the facial aesthetics, which is able to learn the aesthetics attributes automatically from data. By visualizing the attributes in the last fully connected layer, we find that they are largely consistent with the cross-culture coherence theory. Furthermore, the learned attributes in the sallower layer illustrate a potential correlation with the culture specific theory. This research demonstrates a case study of the complicated facial aesthetics cognition.