A novel knowledge distillation approach to create lightweight yet powerful deep learning models for accurate facial landmark detection on embedded systems.