This is only true for multi modal learning, but yeah, in that case we need text and image pairs. More than likely it's possible to pretrain image and language s

This is only true for multi modal learning, but yeah, in that case we need text and image pairs. More than likely it's possible to pretrain image and language separately, and then use a vastly smaller number of pairs. But that's hypothetical.