EgoPHI method details: graph attention over 3D geometry for per-vertex contact and force prediction

cholz · x · 2026-09-10

Author details the EgoPHI method: fusing image features with 3D hand and object geometry, iteratively refining object pose under hand occlusion, modeling intra-mesh geometry and hand-object inter-mesh relations with graph attention plus cross-attention, and predicting per-vertex contact, force magnitude, and 3D direction.

Related event: ETH Zurich's EgoPHI Estimates Hand-Object Contact and 3D Forces from a Single Image(2 posts)→

Original post →

More from Embodied

Embodied channel →