Why does egocentric hand tracking break in the wild?

​A small, closed-door session for about 30 researchers on one question: why does egocentric hand tracking break in the wild?

​We’ll start with HaWoR (CVPR 2025), a recent method for tracking hands in world space from egocentric video, and figure out where it stops working. Hosted by Humyn Labs.

​What we’ll cover

• ​How tracking holds up when hands, or the objects they hold, leave the frame and reappear

• ​Where things break when hands touch objects, or each other

• ​Whether today’s benchmarks measure what robots actually learn

​How it runs

• ​Read the paper and gather your thoughts

• ​Talk it out in small groups, with real cases from your work

• ​Then, an open discussion on the topic

• ​Humyn shares where its own approach breaks, and we learn from each other’s experiences

​Who it’s for

​Researchers working on hand pose, egocentric vision or robot policy learning. If you’ve watched hand tracking fail on real data and have a view on why, we’d value your take. Every seat is approved individually.