LOCATE: Localize and Transfer Object Parts for Weakly Supervised Affordance Grounding

Cross-view-AG+ [32] LOCATE (Ours) GT Figure 1 . State-of-the-art methods in weakly supervised affordance grounding often fail to make accurate predictions for objects with complex structures, e.g., chairs and bicycles. To address this, our model (LOCATE) focuses on localizing and transferring features of object parts, which is able to produce more accurate results.