MAMHOI: Factorizing Scene-Aware Human-Object Interaction through Affordances
MAMHOI splits scene feasibility from motion generation to work around scarce paired human-object-scene data.
The paper introduces an affordance interface: one model predicts where and how an interaction can happen in a 3D indoor scene, then another generates the human-object motion from that affordance. The authors say this lets the system learn scene understanding and interaction dynamics from separate datasets. In experiments, MAMHOI reduces object-scene penetration while preserving interaction quality, producing more physically feasible scene-aware HOI. HF Daily Papers' note
The paper introduces an affordance interface: one model predicts where and how an interaction can happen in a 3D indoor scene, then another generates the human-object motion from that affordance. The authors say this lets the system learn scene understanding and interaction dynamics from separate datasets. In experiments, MAMHOI reduces object-scene penetration while preserving interaction quality, producing more physically feasible scene-aware HOI. HF Daily Papers' note
score 4