FacePaint: An Exploration of Localized Transfer on Facial Expressions

Is it possible to change a person’s facial expression? In this paper, we apply Neural Style Transfer, Image Segmentation, and Generative Adversarial Networks (GANs) to a new application; namely, changing the expression on a human face. Because the human eye is particularly sensitive to distortions of human facial features, accomplishing this goal will require precise and detailed results. We present qualitative results from various architectures, and present the ones that show the most promise with respect to this supervised task. In our experiments, we found that Cycle-GANs show the most promise in this application area. Overall, we present an end-to-end neural framework for realistic expression modification on human faces.