home.social

#imagecaptioning — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #imagecaptioning, aggregated by home.social.

  1. I am looking for a baseline #ImageCaptioning architecture that I can use for a series of quick experiments. I specifically _don't_ want one with a pre-trained image model (the language model can be whatever).

    I would like to train it from scratch on mscoco and ideally it would not take too long to train (but I am not sure how long these usually take tbh). The performance doesn't have to be SOTA ofc, this is just for a side-by-side study.

    Anyone have any good pointers?