Visual Structures Helps Visual Reasoning: Addressing the Binding Problem in VLMs