Search papers, labs, and topics across Lattice.
The paper introduces VLM-3R, a vision-language model that incorporates 3D reconstructive instruction tuning to enable spatial understanding from monocular video. VLM-3R uses a geometry encoder to derive implicit 3D tokens from video frames, which are then aligned with language instructions using a spatial-visual-view fusion technique and a dataset of 200K+ curated QA pairs. The model's performance is evaluated on a newly introduced Vision-Spatial-Temporal Intelligence benchmark, demonstrating its ability to understand temporal 3D context changes and perform robust visual-spatial reasoning.
Unlock human-like spatial reasoning in VLMs with VLM-3R, which reconstructs 3D understanding from monocular video using instruction tuning, bypassing the need for external depth sensors.
The rapid advancement of Large Multimodal Models (LMMs) for 2D images and videos has motivated extending these models to understand 3D scenes, aiming for human-like visual-spatial intelligence. Nevertheless, achieving deep spatial understanding comparable to human capabilities poses significant challenges in model encoding and data acquisition. Existing methods frequently depend on external depth sensors for geometry capture or utilize off-the-shelf algorithms for pre-constructing 3D maps, thereby limiting their scalability, especially with prevalent monocular video inputs and for time-sensitive applications. In this work, we introduce VLM-3R, a unified framework for Vision-Language Models (VLMs) that incorporates 3D Reconstructive instruction tuning. VLM-3R processes monocular video frames by employing a geometry encoder to derive implicit 3D tokens that represent spatial understanding. Leveraging our Spatial-Visual-View Fusion and over 200K curated 3D reconstructive instruction tuning question-answer (QA) pairs, VLM-3R effectively aligns real-world spatial context with language instructions. This enables monocular 3D spatial assistance and embodied reasoning. To facilitate the evaluation of temporal reasoning, we introduce the Vision-Spatial-Temporal Intelligence benchmark, featuring over 138.6K QA pairs across five distinct tasks focused on evolving spatial relationships. Extensive experiments demonstrate that our model, VLM-3R, not only facilitates robust visual-spatial reasoning but also enables the understanding of temporal 3D context changes, excelling in both accuracy and scalability.
