Generation Models Know Space: Unleashing Implicit 3D Priors for Scene Understanding

TL;DR

This paper introduces VEGA-3D, a novel framework that leverages implicit 3D priors learned by video generation models to enhance scene understanding and spatial reasoning in multimodal large language models without explicit 3D data.

Contribution

It proposes a plug-and-play method to extract and utilize implicit 3D priors from pre-trained video diffusion models for improved 3D scene understanding.

Findings

Outperforms state-of-the-art baselines on 3D understanding benchmarks.

Enriches multimodal models with dense geometric cues without explicit 3D supervision.

Demonstrates the effectiveness of generative priors for physical-world reasoning.

Abstract

While Multimodal Large Language Models demonstrate impressive semantic capabilities, they often suffer from spatial blindness, struggling with fine-grained geometric reasoning and physical dynamics. Existing solutions typically rely on explicit 3D modalities or complex geometric scaffolding, which are limited by data scarcity and generalization challenges. In this work, we propose a paradigm shift by leveraging the implicit spatial prior within large-scale video generation models. We posit that to synthesize temporally coherent videos, these models inherently learn robust 3D structural priors and physical laws. We introduce VEGA-3D (Video Extracted Generative Awareness), a plug-and-play framework that repurposes a pre-trained video diffusion model as a Latent World Simulator. By extracting spatiotemporal features from intermediate noise levels and integrating them with semantic representations via a token-level adaptive gated fusion mechanism, we enrich MLLMs with dense geometric cues without explicit 3D supervision. Extensive experiments across 3D scene understanding, spatial reasoning, and embodied manipulation benchmarks demonstrate that our method outperforms state-of-the-art baselines, validating that generative priors provide a scalable foundation for physical-world understanding. Code is publicly available at https://github.com/H-EmbodVis/VEGA-3D.