FYI, they don't actually train end-to-end, they have three separate models that are each trained via self-supervised learning independently (I think, I haven't

FYI, they don't actually train end-to-end, they have three separate models that are each trained via self-supervised learning independently (I think, I haven't read the paper very carefully).