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