Search papers, labs, and topics across Lattice.
3
0
5
1
SOCO reveals that vision models excel in semantic structure but falter in transferring correspondences across categories, exposing a significant gap in multimodal understanding.
Semantic motion anchors enable a dramatic 8.2% boost in text-to-gesture retrieval accuracy by grounding gesture motion in communicative intent, not just kinematics.
Event cameras can now enable significantly more accurate and stable egocentric 3D human pose estimation, thanks to a novel state machine approach that directly leverages fine-grained event dynamics.
