Using our detection framework to obtain an annotated dataset. we can calculate the artist mean facial shape μ artist (similar to artist distributions), and create an average portrait representing the different artists. To create an average portrait, we simply warp the collection of annotated portraits to the artists mean shape μ artist and calculate the mean RGB values of the warped portraits collection.
Here we show several average portraits of different artists. Using these images we gain some interesting insights regarding the artist's style preferences: color schemes (warm colors for Modigliani, cool colors for Keane), colorfulness (saturated colors for Party, dull colors for Foujita), abstraction level (abstract portraits for Leger and Chagall, realistic, detailed portraits for Rafael and Rembrandt), textures (smooth portraits for Party and Hindu Gods, rough textured portraits for Bratby and Kokoschka), emotions (anger in Comics, happiness in Disney), and gender (male for Rembrandt, female for Lichtenstein).