Gaurav Kaul

We present Global Matching Random Walks, a simple self-supervised approach to the Tracking Any Point (TAP) problem, accepted to #ECCV2024. We train a global matching transformer to find cycle consistent tracks through video via contrastive random walks (CRW).

(6/6)Huge thanks to @UM_MICDE for supporting this work with the catalyst grant, @OpenBehavior for creating an open source animal behavior video repository ~ along with all the scientist who uploaded their videos, and @huggingface for hosting our datasets!

(5/6) All in all, Detect Any Mouse Model, or DAMM for short, is great at detecting mice out of the box and can quickly adapt to new settings with just a few training points! Check out how you can use DAMM in your projects via our GitHub, and explore our datasets on Hugging Face.

Detect Any Mouse Model accepted into Scientific Reports! Paper: nature.com/articles/s4159… Project page: web.eecs.umich.edu/gkaul/DAMM/ GitHub: github.com/backprop64/DAMM Datasets: huggingface.co/datasets/gaura… W/ Jonathan McDevitt, Justin Johnson (@jcjohnss), Ada Eban-Rothschild (🧵 below 1/6)