Cog3DMap: Multi-View Vision-Language Reasoning with 3D Cognitive Maps

TL;DR

Cog3DMap enhances multi-view vision-language models by constructing an explicit 3D memory, grounding tokens in 3D space for improved spatial reasoning, and achieving state-of-the-art results on relevant benchmarks.

Contribution

The paper introduces Cog3DMap, a novel framework that explicitly constructs a 3D memory from multi-view images for better spatial reasoning in vision-language models.

Findings

Achieves state-of-the-art performance on spatial reasoning benchmarks.

Constructs explicit 3D tokens with semantic and geometric information.

Enables direct reasoning over a structured 3D map.

Abstract

Precise spatial understanding from multi-view images remains a fundamental challenge for Multimodal Large Language Models (MLLMs), as their visual representations are predominantly semantic and lack explicit geometric grounding. While existing approaches augment visual tokens with geometric cues from visual geometry models, their MLLM is still required to implicitly infer the underlying 3D structure of the scene from these augmented tokens, limiting its spatial reasoning capability. To address this issue, we introduce Cog3DMap, a framework that recurrently constructs an explicit 3D memory from multi-view images, where each token is grounded in 3D space and possesses both semantic and geometric information. By feeding these tokens into the MLLM, our framework enables direct reasoning over a spatially structured 3D map, achieving state-of-the-art performance on various spatial reasoning benchmarks. Code will be made publicly available.