Exploring the performance and robustness of Vision foundation models