A Survey of Vision-Language Pre-Trained Models