Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces

Chen Yang, Guanxin Lin, Youquan He, Peiyao Chen, Guanghe Liu, Yufan Mo, Zhouyuan Xu, Linhao Wang, Guohui Zhang, Zihang Zhang, Shenxiang Zeng, Chen Wang, Jiansheng Fan
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:145388-145423, 2026.

Abstract

Spatial intelligence is crucial for vision–language models (VLMs), yet many scene-centric benchmarks evaluate unconstrained environments where a single image may admit multiple plausible 3D interpretations. We introduce SSI-Bench, a VQA benchmark for Structure-Centric Spatial Reasoning (SCSR) in constraint-governed spaces. Built from complex real-world 3D structures, it uses structural constraints from geometry, topology, and physical feasibility to make component relations more determinate from visual evidence. The benchmark contains 1,000 ranking questions spanning geometric and topological reasoning, where correct ordering requires resolving all candidate-wise 3D relations, imposing stronger demands on spatial understanding. It is created through a fully human-centered pipeline with over 400 researcher-hours of image curation, component annotation, and question design. Evaluating 31 VLMs reveals a large gap to humans: the best open-source model achieves 22.2% accuracy and the strongest closed-source model reaches 33.6%, while humans score 91.6%. Further results show that chain-of-thought reasoning brings only marginal gains, and error analysis reveals fundamental limitations in current models’ spatial understanding within constraint-governed spaces. Project page: https://ssi-bench.github.io.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-yang26ar, title = {Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces}, author = {Yang, Chen and Lin, Guanxin and He, Youquan and Chen, Peiyao and Liu, Guanghe and Mo, Yufan and Xu, Zhouyuan and Wang, Linhao and Zhang, Guohui and Zhang, Zihang and Zeng, Shenxiang and Wang, Chen and Fan, Jiansheng}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {145388--145423}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/yang26ar/yang26ar.pdf}, url = {https://proceedings.mlr.press/v306/yang26ar.html}, abstract = {Spatial intelligence is crucial for vision–language models (VLMs), yet many scene-centric benchmarks evaluate unconstrained environments where a single image may admit multiple plausible 3D interpretations. We introduce SSI-Bench, a VQA benchmark for Structure-Centric Spatial Reasoning (SCSR) in constraint-governed spaces. Built from complex real-world 3D structures, it uses structural constraints from geometry, topology, and physical feasibility to make component relations more determinate from visual evidence. The benchmark contains 1,000 ranking questions spanning geometric and topological reasoning, where correct ordering requires resolving all candidate-wise 3D relations, imposing stronger demands on spatial understanding. It is created through a fully human-centered pipeline with over 400 researcher-hours of image curation, component annotation, and question design. Evaluating 31 VLMs reveals a large gap to humans: the best open-source model achieves 22.2% accuracy and the strongest closed-source model reaches 33.6%, while humans score 91.6%. Further results show that chain-of-thought reasoning brings only marginal gains, and error analysis reveals fundamental limitations in current models’ spatial understanding within constraint-governed spaces. Project page: https://ssi-bench.github.io.} }
Endnote
%0 Conference Paper %T Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces %A Chen Yang %A Guanxin Lin %A Youquan He %A Peiyao Chen %A Guanghe Liu %A Yufan Mo %A Zhouyuan Xu %A Linhao Wang %A Guohui Zhang %A Zihang Zhang %A Shenxiang Zeng %A Chen Wang %A Jiansheng Fan %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-yang26ar %I PMLR %P 145388--145423 %U https://proceedings.mlr.press/v306/yang26ar.html %V 306 %X Spatial intelligence is crucial for vision–language models (VLMs), yet many scene-centric benchmarks evaluate unconstrained environments where a single image may admit multiple plausible 3D interpretations. We introduce SSI-Bench, a VQA benchmark for Structure-Centric Spatial Reasoning (SCSR) in constraint-governed spaces. Built from complex real-world 3D structures, it uses structural constraints from geometry, topology, and physical feasibility to make component relations more determinate from visual evidence. The benchmark contains 1,000 ranking questions spanning geometric and topological reasoning, where correct ordering requires resolving all candidate-wise 3D relations, imposing stronger demands on spatial understanding. It is created through a fully human-centered pipeline with over 400 researcher-hours of image curation, component annotation, and question design. Evaluating 31 VLMs reveals a large gap to humans: the best open-source model achieves 22.2% accuracy and the strongest closed-source model reaches 33.6%, while humans score 91.6%. Further results show that chain-of-thought reasoning brings only marginal gains, and error analysis reveals fundamental limitations in current models’ spatial understanding within constraint-governed spaces. Project page: https://ssi-bench.github.io.
APA
Yang, C., Lin, G., He, Y., Chen, P., Liu, G., Mo, Y., Xu, Z., Wang, L., Zhang, G., Zhang, Z., Zeng, S., Wang, C. & Fan, J.. (2026). Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:145388-145423 Available from https://proceedings.mlr.press/v306/yang26ar.html.

Related Material