metadata

task_categories:
  - other
pretty_name: SURPRISE3D
tags:
  - 3d
  - spatial-reasoning
  - segmentation
  - vision-language
  - embodied-ai
library_name: datasets
license: mit
size_categories:
  - 100K<n<1M
language:
  - en

SURPRISE3D Dataset

📄 Paper: SURPRISE3D: A Dataset for Spatial Understanding and Reasoning in Complex 3D Scenes

🔗 arXiv: arxiv:2507.07781

💻 Code: GitHub Repository

Dataset Description

SURPRISE3D is a novel dataset designed to evaluate language-guided spatial reasoning segmentation in complex 3D scenes. As detailed in our paper, this dataset addresses the critical gap in current 3D vision-language research where existing datasets often mix semantic cues with spatial context.

Key Features:

200k+ vision-language pairs across 900+ detailed indoor scenes from ScanNet++ v2
2.8k+ unique object classes
89k+ human-annotated spatial queries crafted without object names to mitigate shortcut biases
Comprehensive coverage of spatial reasoning skills including:
- Relative position reasoning
- Narrative perspective understanding
- Parametric perspective analysis
- Absolute distance reasoning

Citation

If you use SURPRISE3D in your research, please cite our paper:

@article{huang2025surprise3d,
  title={SURPRISE3D: A Dataset for Spatial Understanding and Reasoning in Complex 3D Scenes},
  author={Huang, Jiaxin and Li, Ziwen and Zhang, Hanlve and Chen, Runnan and He, Xiao and Guo, Yandong and Wang, Wenping and Liu, Tongliang and Gong, Mingming},
  journal={arXiv preprint arXiv:2507.07781},
  year={2025}
}