Sep 05, 2025 Object-Guided Visual Tokens: Eliciting Compositional Reasoning in Multimodal Language Models