Self-supervised vision transformer-based few-shot learning for facial expression recognition.