3D-Aware VLMs with Implicit and Explicit Geometries

Summary

Despite rapid progress, most existing vision-language models (VLMs) built from 2D visual inputs often struggle when handling various 3D tasks that require fine-grained spatial understanding and reasoning.