Separating head position and orientation from facial expression features to improve the model's focus on meaningful deformations.