Limited Data, Unlimited Potential: A Study on ViTs Augmented by Masked Autoencoders