|
Ben Schmidt@bschmidt.bsky.social |
Announcing a new open-weights model from Nomic -- it puts images into the same space as our best-in-class text embedder. blog.nomic.ai/posts/nomic-.... That means you can do rich language search directly on any image dataset! E.g., searching for 'kitschy Americana' at the Met gets cigarette ads.
2 replies 3 reposts 18 likes
|
|
Ben Schmidt
@bschmidt.bsky.social
[ View ] |
We designed our own map for the launch with photos, but some undergrad social media influencer made a much better one so I'll link that: all (?) the images in the Metropolitan Museum's online collection embedded, and text queryable.
atlas.nomic.ai/data/andrewg...
1 replies 0 reposts 2 likes
|
Thomas Dickerson
@elfprince13.mumak.app
[ View ] |
kinda interesting that forcing the image encoder to use the existing text embedding space outperformed training them jointly. any intuition what's going on with that?
1 replies 0 reposts 1 likes