Enregistré dans:
Détails bibliographiques
Auteurs principaux: Jeong, Suchae, Song, Jaehwi, Lee, Haeone, Kim, Hanna, Kim, Jian, Lee, Dongjun, Shin, Dong Kyu, Kim, Changyeon, Hahm, Dongyoon, Jin, Woogyeol, Choi, Juheon, Lee, Kimin
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2603.27967
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866910084372627456
author Jeong, Suchae
Song, Jaehwi
Lee, Haeone
Kim, Hanna
Kim, Jian
Lee, Dongjun
Shin, Dong Kyu
Kim, Changyeon
Hahm, Dongyoon
Jin, Woogyeol
Choi, Juheon
Lee, Kimin
author_facet Jeong, Suchae
Song, Jaehwi
Lee, Haeone
Kim, Hanna
Kim, Jian
Lee, Dongjun
Shin, Dong Kyu
Kim, Changyeon
Hahm, Dongyoon
Jin, Woogyeol
Choi, Juheon
Lee, Kimin
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2603_27967
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Learning Multi-View Spatial Reasoning from Cross-View Relations
Jeong, Suchae
Song, Jaehwi
Lee, Haeone
Kim, Hanna
Kim, Jian
Lee, Dongjun
Shin, Dong Kyu
Kim, Changyeon
Hahm, Dongyoon
Jin, Woogyeol
Choi, Juheon
Lee, Kimin
Computer Vision and Pattern Recognition
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.
title Learning Multi-View Spatial Reasoning from Cross-View Relations
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.27967