Perception-driven facial expression synthesis

► We propose a platform to create meaningful facial expressions without performance data. ► We first synthesize random animations by modulating Action Units (AUs) taken from the FACS. ► We present these to human observers and use the random noise to model the internal representations.