VLM3: Vision Language Models Are Native 3D Learners

2026-05-28

AI Summary

This work demonstrates for the first time that standard vision-language models inherently possess native 3D learning capabilities, achieving competitive multi-task 3D understanding without any architectural modifications. By introducing focal-length normalization, text-guided pixel localization, and training on large-scale mixed datasets, the proposed approach matches or surpasses specialized expert models across diverse tasks—including depth estimation (improving accuracy from 0.84 to 0.90), pixel correspondence, camera pose estimation, and object-level 3D understanding. The method establishes a new, unified paradigm for 3D perception that is both simple and scalable, overcoming the limitations of task-specific architectures and enabling seamless integration with general-purpose vision-language systems.

Abstract

Vision Language Models (VLMs) enable a unified model to solve various vision tasks through prompting. They have shown promising performance in semantic understanding. However, 3D understanding still largely relies on expert vision models with complex task-specific designs. The key argument this work wants to make is that VLMs are native 3D learners. Our in-depth large scale study shows that 1) focal length unification, 2) text-based pixel reference and 3) data mixture and scaling, are all you need for effective 3D learning. Model architecture changes, large models, heavy data augmentations, and complex losses including the regression formulation, many of which form the foundation of expert vision models, are actually not necessary conditions. As a result, we propose VLM3, a scalable method with the simplest design that enables standard VLMs to master diverse 3D tasks. VLM3 not only advances the VLM depth estimation accuracy by a large margin (0.84 -> 0.9), but also enables diverse 3D tasks such as pixel correspondence, camera pose estimation and object-level 3D understanding, matching expert vision model accuracy while maintaining standard architectures and text-based training. We believe VLM3 opens up a new paradigm for simple and scalable 3D learning.

Problem

• Vision Language Models
• 3D understanding
• depth estimation
• camera pose estimation
• pixel correspondence

Innovation

• Vision Language Models
• 3D understanding
• focal length unification
• text-based pixel reference
• scalable 3D learning