aiqbench analysis
GraFT addresses the critical limitation of current multimodal large language models in 3D spatial reasoning by introducing a training-free framework that leverages compact 3D scene graphs. This approach eliminates the need for costly supervised fine-tuning or dedicated geometry encoders, offering deterministic geometric measurements and robust viewpoint transformations directly through symbolic tools and BEV rendering. The results demonstrate significant performance gains on standard benchmarks without altering the underlying model backbone, marking a notable step toward efficient spatial intelligence in AI systems.
GraFT: A Training-Free Framework for Spatial Reasoning in Multimodal Large Language Models via 3D Scene Graphs
arXiv:2609.03892v1 Announce Type: cross Abstract: 3D spatial reasoning underpins understanding and acting in the physical world, yet it remains unreliable in current multimodal large language models (MLLMs). These models falter at precise geometric measurement, at transforming between egocentric and allocentric viewpoints, and at grounding fine-grained appearance. The most common remedies fine-tune the model on large-scale curated spatial-reasoning datasets or attach dedicated encoders for 3D geometry, which typically couples the solution to costly supervision and a specific backbone. We instead introduce GraFT, a training-free framework that supplies the missing 3D structure through a compact, easily maintained 3D scene graph (3DSG). From this 3DSG, GraFT provides three spatial reasoning capabilities: (1) deterministic geometry through symbolic tools, (2) allocentric layout through a bird's-eye-view (BEV) rendering, and (3) visual-attribute grounding through task-relevant egocentric frames. On ScanQA, GraFT improves every metric over the same-backbone baseline, raising CIDEr by 27%. On VSI-Bench, GraFT improves frozen MLLMs by up to 65%, surpassing every proprietary and general-purpose open-source baseline, and several prominent fine-tuned spatial models.
Abstract:3D spatial reasoning underpins understanding and acting in the physical world, yet it remains unreliable in current multimodal large language models (MLLMs). These models falter at precise geometric measurement, at transforming between egocentric and allocentric viewpoints, and at grounding fine-grained appearance. The most common remedies fine-tune the model on large-scale curated spatial-reasoning datasets or attach dedicated encoders for 3D geometry, which typically couples the solution to costly supervision and a specific backbone. We instead introduce GraFT, a training-free framework that supplies the missing 3D structure through a compact, easily maintained 3D scene graph (3DSG). From this 3DSG, GraFT provides three spatial reasoning capabilities: (1) deterministic geometry through symbolic tools, (2) allocentric layout through a bird's-eye-view (BEV) rendering, and (3) visual-attribute grounding through task-relevant egocentric frames. On ScanQA, GraFT improves every metric over the same-backbone baseline, raising CIDEr by 27%. On VSI-Bench, GraFT improves frozen MLLMs by up to 65%, surpassing every proprietary and general-purpose open-source baseline, and several prominent fine-tuned spatial models.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
arXiv:2609.03892 [cs.CV]
(or arXiv:2609.03892v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.03892 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 264 words
Related coverage
Trace, Verify, and Correct: A Training-Free Framework for Spatial Reasoning in Multimodal LLMs
arXiv · Aug 6, 2026
Beyond RGB: Benchmarking and Enhancing MLLMs for Hyperspectral Image Understanding via Training-Free Reasoning Framework
arXiv · Aug 25, 2026
Spatial-IQ: Deconstructing Spatial Intelligence via Hierarchical Capability Tests
arXiv · Jul 28, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026