Learning Multi-View Spatial Reasoning from Cross-View Relations

TL;DR

This paper introduces Cross-View Relations (XVR), a large-scale dataset for training vision-language models in multi-view spatial reasoning, significantly improving their performance on related benchmarks and robotic tasks.

Contribution

The creation of XVR, a comprehensive dataset for multi-view spatial reasoning, and demonstrating its effectiveness in enhancing vision-language models for embodied AI applications.

Findings

VLMs fine-tuned on XVR outperform existing models on multi-view reasoning benchmarks.

XVR-trained models show improved success rates in robotic manipulation tasks.

Explicit training on cross-view relations enhances transfer to real-world robotic applications.

Abstract

Vision-language models (VLMs) have achieved impressive results on single-view vision tasks, but lack the multi-view spatial reasoning capabilities essential for embodied AI systems to understand 3D environments and manipulate objects across different viewpoints. In this work, we introduce Cross-View Relations (XVR), a large-scale dataset designed to teach VLMs spatial reasoning across multiple views. XVR comprises 100K vision-question-answer samples derived from 18K diverse 3D scenes and 70K robotic manipulation trajectories, spanning three fundamental spatial reasoning tasks: Correspondence (matching objects across views), Verification (validating spatial relationships), and Localization (identifying object positions). VLMs fine-tuned on XVR achieve substantial improvements on established multi-view and robotic spatial reasoning benchmarks (MindCube and RoboSpatial). When integrated as backbones in Vision-Language-Action models, XVR-trained representations improve success rates on RoboCasa. Our results demonstrate that explicit training on cross-view spatial relations significantly enhances multi-view reasoning and transfers effectively to real-world robotic manipulation.