{"ID":2868905,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2509.16087","arxiv_id":"2509.16087","title":"See\u0026Trek: Training-Free Spatial Prompting for Multimodal Large Language Model","abstract":"We introduce SEE\u0026TREK, the first training-free prompting framework tailored to enhance the spatial understanding of Multimodal Large Language Models (MLLMS) under vision-only constraints. While prior efforts have incorporated modalities like depth or point clouds to improve spatial reasoning, purely visualspatial understanding remains underexplored. SEE\u0026TREK addresses this gap by focusing on two core principles: increasing visual diversity and motion reconstruction. For visual diversity, we conduct Maximum Semantic Richness Sampling, which employs an off-the-shell perception model to extract semantically rich keyframes that capture scene structure. For motion reconstruction, we simulate visual trajectories and encode relative spatial positions into keyframes to preserve both spatial relations and temporal coherence. Our method is training\u0026GPU-free, requiring only a single forward pass, and can be seamlessly integrated into existing MLLM'S. Extensive experiments on the VSI-B ENCH and STI-B ENCH show that S EE \u0026T REK consistently boosts various MLLM S performance across diverse spatial reasoning tasks with the most +3.5% improvement, offering a promising path toward stronger spatial intelligence.","short_abstract":"We introduce SEE\u0026TREK, the first training-free prompting framework tailored to enhance the spatial understanding of Multimodal Large Language Models (MLLMS) under vision-only constraints. While prior efforts have incorporated modalities like depth or point clouds to improve spatial reasoning, purely visualspatial under...","url_abs":"https://arxiv.org/abs/2509.16087","url_pdf":"https://arxiv.org/pdf/2509.16087v1","authors":"[\"Pengteng Li\",\"Pinhao Song\",\"Wuyang Li\",\"Weiyu Guo\",\"Huizai Yao\",\"Yijie Xu\",\"Dugang Liu\",\"Hui Xiong\"]","published":"2025-09-19T15:30:26Z","proceeding":"cs.CV","tasks":"[\"cs.CV\",\"cs.AI\"]","methods":"[\"Large Language Model\",\"Language Model\"]","has_code":false}