Describir: Improving Laryngoscopy Image Analysis Through Integration of Global Information and Local Features in VoFoCD Dataset.