Abstract: Human-Object Interaction (HOI) detection remains challenging due to the semantic ambiguity of interaction categories and the limited discriminability of their feature representations.