MAMHOI: Factorizing Scene-Aware Human-Object Interaction through Affordances
MAMHOI splits the problem into affordance prediction first, motion generation second.
The paper says paired human-object-scene training data is scarce, so it learns scene feasibility and interaction dynamics from separate supervision sources. A scene-conditioned model predicts where and how an interaction can happen, then an affordance-conditioned HOI model generates the human-object motion. In indoor-scene experiments, the authors report less object-scene penetration while maintaining interaction quality. ArXiv · AI/CL/LG's note
The paper says paired human-object-scene training data is scarce, so it learns scene feasibility and interaction dynamics from separate supervision sources. A scene-conditioned model predicts where and how an interaction can happen, then an affordance-conditioned HOI model generates the human-object motion. In indoor-scene experiments, the authors report less object-scene penetration while maintaining interaction quality. ArXiv · AI/CL/LG's note
score 4