Tanishq Mathew Abraham, Ph.D.#359
What if the metadata from acquisition was enough? 👀
Sometimes the most useful labels aren't labels at all.
Introducing📄 Who Needs Labels? Adapting Vision Foundation Models With the Metadata You Already Have
Tanishq Mathew Abraham, Ph.D.#359What if the metadata from acquisition was enough? 👀
Sometimes the most useful labels aren't labels at all.
Introducing📄 Who Needs Labels? Adapting Vision Foundation Models With the Metadata You Already Have
Users expressed excitement about the new method for adapting vision models using acquisition metadata alone because it looks innovative and promising.
Say hello to FINO: a label-free way to adapt foundation models using metadata alone.
🔬 Scientific datasets come with rich metadata almost for free: antibodies, microscope plates, camera traps, locations, timestamps, imaging views, and more.
See @ https://arxiv.org/pdf/2606.05107
@hassonofer_ We're working on it but yes - there will be one. I'll update when it's out and dm you if you're interested
ETA 1~2 weeks
Thanks to my co-authors, @seungeunyi10, @KartikAhuja1, @TheoMoutakanni, @huyvvo, @p_bojanowski, Wolfgang M. Pernice, Camille Couprie and @loiclandrieu