Round 1 of the MABe Challenge is over, with winner announcements coming soon- but in the meantime, Round 2 has begun! Your goal is still to produce meaningful unsupervised embeddings of animals' actions- but now, you have the raw videos to work with. 1/5

17 38

The basic idea here is to train MLPs to predict a CLIP image embedding from a text embedding, and vice versa; then normalize and add in the embeddings.

0 7

that reminds me. got a color suggestion for the beddings beneath this? here's some suggestions. but ideas would be welcome.

1 6

Me reading a bunch of sh*t I don't understand: "I'm creating embeddings"

0 2

Averaging CLIP embeddings over many permutations of the form "a photograph of X", "X trending on artstation", "X rendered in Blender/Unity," etc produces fairly dramatic improvements in some cases.

e.g. "Jabba the Hutt smoking a cigar"

Left: single text, Right: avg many texts

12 86

Man I woke up today with some random allergic reaction so there's hives all over my torso. Ugh. I hate it when this happens.
Changed all my beddings + clothes + took a bath to help it calm down. My skin randomly decides to do this every few years. Sigh. 🙃

3 61