A Vision-Language Model with An Ensemble of Experts