📦 Spatial Perception And Reasoning Dataset (SPAR-7M)

A large-scale vision-language dataset designed for spatial perception and reasoning.

SPAR-7M contains over 7 million QA pairs across 33 diverse spatial tasks, generated from 4,500+ richly annotated 3D indoor scenes. It supports single-view, multi-view, and video-based image inputs, and features both perception and reasoning-oriented question types.

This dataset serves as the foundation for SPAR-Bench, and is suitable for pretraining, multitask learning, and spatial grounding research.

This version supports single-view, multi-view, and video-based inputs.

📥 Download

We provide two versions of the dataset:

Version	Description
SPAR-7M	RGB-only images + QA annotations
SPAR-7M-RGBD	Includes depths, camera intrinsics, and pose matrices for 3D-aware training

You can download both versions from Hugging Face:

# Download SPAR-7M (default)
huggingface-cli download jasonzhango/SPAR-7M --repo-type dataset

# Download SPAR-7M-RGBD (with depth and camera parameters)
huggingface-cli download jasonzhango/SPAR-7M-RGBD --repo-type dataset

These datasets are split into multiple .tar.gz parts due to Hugging Face file size limits. After downloading all parts, run the following to extract:

# For SPAR-7M
cat spar-*.tar.gz | tar -xvzf -

# For SPAR-7M-RGBD
cat spar-rgbd-*.tar.gz | tar -xvzf -

Alternatively, if Hugging Face is not accessible, you can use the provided script:

wget https://hf-mirror.com/hfd/hfd.sh

chmod a+x hfd.sh

export HF_ENDPOINT=https://hf-mirror.com

./hfd.sh jasonzhango/SPAR-7M --dataset
./hfd.sh jasonzhango/SPAR-7M-RGBD --dataset

The dataset directory structure is:

spar/
├── rxr/
├── scannet/
│   ├── images/
│   |   └── scene0000_00/
│   |       ├── image_color/
│   |       ├── video_color/
│   |       ├── image_depth/           # only in SPAR-7M-RGBD
│   |       ├── video_depth/           # only in SPAR-7M-RGBD
│   |       ├── pose/                  # only in SPAR-7M-RGBD
│   |       ├── video_pose/            # only in SPAR-7M-RGBD
│   |       ├── intrinsic/             # only in SPAR-7M-RGBD
│   |       └── video_idx.txt
│   └── qa_jsonl/
│       ├── train/
│       |   ├── depth_prediction_oo/
│       |   |   ├── fill/
│       |   |   |   └── fill_76837.jsonl
│       |   |   ├── select/
│       |   |   └── sentence/
│       |   ├── obj_spatial_relation_oc/
│       |   └── spatial_imagination_oo_mv/
│       └── val/
├── scannetpp/
└── structured3d/

Each QA task (e.g., depth_prediction_oc, spatial_relation_oo_mv, etc.) is organized by task type, with subfolders for different answer formats:

• fill/ — numerical or descriptive answers
• select/ — multiple choice
• sentence/ — natural language answers

📚 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 Xu, Yueming and Huang, Ze and Mei, Jilin and Chen, Junhui and Zhou, Yanpeng 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},
}