3DRS: MLLMs Need 3D-Aware Representation Supervision for Scene Understanding

TL;DR

This paper introduces 3DRS, a framework that improves multimodal large language models' 3D scene understanding by incorporating supervision from pretrained 3D models, leading to better performance on various benchmarks.

Contribution

The paper proposes a novel 3D-aware supervision method for MLLMs, enhancing their 3D representation capabilities by distilling knowledge from 3D foundation models.

Findings

Improved performance on visual grounding, captioning, and question answering tasks.

Strong correlation between 3D-awareness and downstream task success.

Consistent gains across multiple MLLMs and benchmarks.

Abstract

Recent advances in scene understanding have leveraged multimodal large language models (MLLMs) for 3D reasoning by capitalizing on their strong 2D pretraining. However, the lack of explicit 3D data during MLLM pretraining limits 3D representation capability. In this paper, we investigate the 3D-awareness of MLLMs by evaluating multi-view correspondence and reveal a strong positive correlation between the quality of 3D-aware representation and downstream task performance. Motivated by this, we propose 3DRS, a framework that enhances MLLM 3D representation learning by introducing supervision from pretrained 3D foundation models. Our approach aligns MLLM visual features with rich 3D knowledge distilled from 3D models, effectively improving scene understanding. Extensive experiments across multiple benchmarks and MLLMs -- including visual grounding, captioning, and question answering -- demonstrate consistent performance gains. Project page: https://visual-ai.github.io/3drs