A Style-Based Caricature Generator

Lamyanba Laishram, Muhammad Shaheryar, Jong Taek Lee, Soon Ki Jung

Abstract: A facial caricature is a creation of new artistic and exaggerated faces which translates into a real image to convey sarcasm or humor while keeping the identity of the subject. In this work, we proposed a new way to create caricatures by exaggerating facial features like the eyes and mouth while keeping the facial contour intact and a realistic style. Our method can be categorized into two steps. First, the facial exaggeration process transformed faces into caricature face images while maintaining facial contours. Second, the appearance style generator is trained in unpaired using the generated caricature faces to produce a facial caricature that can change to any realistic style of our preference. Experimental results show our model produces more realistic and disentangled caricature images as compared to some of the previous methods. Our method can also generate caricature images from real images.