Preprint
Aug 2026
Chain of Spatial Thoughts: Modality-Agnostic Spatial Grounding for Vision Language Models
Space Tokens is introduced, a lightweight, architecture-agnostic framework that equips VLMs with explicit continuous spatial representations without requiring additional inference-time modules, and demonstrates that continuous spatial tokens provide an effective, interpretable, and computationally efficient mechanism for integrating geometric reasoning into large vision-language models.
Hunter Schofield, Mohammed Elmahgiubi, Mohammad Mahdavian et al.
· 0 citations