🎯 SPAR-Bench-RGBD
A depth-enhanced version of SPAR-Bench for evaluating 3D-aware spatial reasoning in vision-language models.
SPAR-Bench-RGBD extends the full SPAR-Bench with additional depths, camera intrinsics, and pose information, enabling evaluation of models with geometric or 3D-awareness capabilities.
The benchmark contains 7,207 manually verified QA pairs across 20 spatial tasks and supports single-view and multi-view inputs.
📥 Load with datasets
from datasets import load_dataset
spar_rgbd = load_dataset("jasonzhango/SPAR-Bench-RGBD")
🕹️ Evaluation
SPAR-Bench-RGBD 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},
}