Face modeling has been paid much attention in the field of visual computing.There exist many scenarios, including cartoon characters, avatars for socialmedia, 3D face caricatures as well as face-related art and design, wherelow-cost interactive face modeling is a popular approach especially amongamateur users. In this paper, we propose a deep learning based sketching systemfor 3D face and caricature modeling. This system has a labor-efficientsketching interface, that allows the user to draw freehand imprecise yetexpressive 2D lines representing the contours of facial features. A novel CNNbased deep regression network is designed for inferring 3D face models from 2Dsketches. Our network fuses both CNN and shape based features of the inputsketch, and has two independent branches of fully connected layers generatingindependent subsets of coefficients for a bilinear face representation. Oursystem also supports gesture based interactions for users to further manipulateinitial face models. Both user studies and numerical results indicate that oursketching system can help users create face models quickly and effectively. Asignificantly expanded face database with diverse identities, expressions andlevels of exaggeration is constructed to promote further research andevaluation of face modeling techniques.
展开▼
