VLM-3R is a unified Vision-Language Model (VLM) framework integrating 3D reconstructive instruction tuning for deep spatial understanding from monocular video. The rapid advancement of Large ...