This work presents Fed-MP, a novel open-vocabulary federated learning framework that leverages pre-trained vision-language models to enable accurate predictions for queries involving unseen classes.
