The abstract from the paper is the following:
While the general idea of self-supervised learning is identical across modalities, the actual algorithms and
objectives differ widely because they were developed with a single modality in mind.