🎯 SPAR-Bench-Tiny

A lightweight subset of SPAR-Bench for fast evaluation of spatial reasoning in vision-language models (VLMs).

SPAR-Bench-Tiny contains 1,000 manually verified QA pairs — 50 samples per task across 20 spatial tasks — covering single-view and multi-view inputs.

This dataset mirrors the structure and annotation of the full SPAR-Bench, but is 10× smaller, making it ideal for low-latency evaluation.

📥 Load with datasets

from datasets import load_dataset
spar_tiny = load_dataset("jasonzhango/SPAR-Bench-Tiny")

🕹️ Evaluation

SPAR-Bench-Tiny uses the same evaluation protocol and metrics as the full SPAR-Bench.

We provide an evaluation pipeline in our GitHub repository, built on top of lmms-eval.

📚 Bibtex

If you find this project or dataset helpful, please consider citing our paper:

@article{zhang2025from,
    title={From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D},
    author={Zhang, Jiahui and Chen, Yurui and Zhou, Yanpeng and Xu, Yueming and Huang, Ze and Mei, Jilin and Chen, Junhui and Yuan, Yujie and Cai, Xinyue and Huang, Guowei and Quan, Xingyue and Xu, Hang and Zhang, Li},
    year={2025},
    journal={arXiv preprint arXiv:2503.22976},
}