Facile: a System for Artificial Face Dataset Generation

Abstract

We present Facile, a system for the generation of artificial face datasets aimed to assist the training and evaluation phases of face related algorithms. Facile is able to generate, starting from a set of textured 3D faces with neutral expressions, a new dataset containing synthesized facial expressions, multiple viewpoint positions and lighting conditions. The system is featured by a facial animation algorithm based on a mass-spring system. Muscles, defined on a reference face, modify springs contractions when activated. Dense correspondence with the reference face is computed for each input face: in this way muscles are automatically positioned and the desired facial expression is generated through physical simulation. Output images can be rendered with different lighting conditions and viewpoint positions depending on the task requirements.