Using Self-Supervised Auxiliary Tasks to Improve Fine-Grained Facial Representation