Pokerme/view2space-v1

VIEW2SPACE v1

VIEW2SPACE v1 is a multi-view vision-language evaluation dataset for spatial reasoning. [image]

Associated paper:

• —VIEW2SPACE: Studying Multi-View Visual Reasoning from Sparse Observations - ECCV 2026 🚀
• —arXiv: 2603.16506
• —Project Page: Project Page

Related VIEW2SPACE Releases

• —Training release: Pokerme/view2space-train
• —4B model checkpoint: Pokerme/view2space_4b
• —Collection: Pokerme/view2space

The public release is organized into three subsets:

• —count
• —detect
• —mcq

For official code, updates, and usage instructions, please see:

GitHub💻: VIEW2SPACE GitHub repository

Directory Structure

view2space-v1-release/
  README.md
  count/
    overall.jsonl
  detect/
    overall.jsonl
  mcq/
    overall.jsonl
  images/
    img_000001.png
    img_000002.png
    ...

Each subset contains one overall.jsonl file. Each line is one question.

Data Format

Each JSONL record contains the following public fields:

• —q_idx: question id, such as mcq_000001
• —q_type: question family, one of mcq, detect, or count
• —question: question text
• —options: multiple-choice options for MCQ questions; empty for non-MCQ questions
• —question_prompt: extra prompt text when applicable
• —answer: ground-truth answer
• —image_paths: relative paths to the images used by the question
• —supporting.draw_boxes: optional input boxes associated with one or more images

Notes:

• —image_paths are relative to the dataset root, not to the subset folder.
• —Example image path: images/img_000123.png
• —q_type uses only the public family labels mcq, detect, and count.

Example Record

{
  "q_idx": "mcq_000001",
  "q_type": "mcq",
  "question": "All views are captured from the same static scene arrangement, but from different positions and angles.\nIn view 2, consider the nearest 'man with an orange hat, crouching'. Where is the 'white temple' relative to this 'man with an orange hat, crouching'?",
  "options": {
    "A": "left",
    "B": "right",
    "C": "front",
    "D": "back"
  },
  "question_prompt": "",
  "answer": "C",
  "image_paths": [
    "images/img_000813.png",
    "images/img_000806.png",
    "images/img_000815.png"
  ],
  "supporting": {
    "draw_boxes": null
  }
}

Basic Usage

import json
from pathlib import Path

dataset_root = Path("/path/to/view2space-v1-release")
jsonl_path = dataset_root / "mcq" / "overall.jsonl"

with jsonl_path.open("r", encoding="utf-8") as f:
    first = json.loads(next(f))

image_files = [dataset_root / rel_path for rel_path in first["image_paths"]]
print(first["q_idx"])
print(first["q_type"])
print(image_files)

Citation

License and Citation

This dataset is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Attribution is required under the license. For academic and research use, citation in the resulting paper, report, model card, dataset card, or public documentation is expected. Please cite this dataset whenever it is used for training, evaluation, benchmarking, data analysis, or as part of a larger dataset mixture.

@article{ke2026view2space,
  title={VIEW2SPACE: Studying Multi-View Visual Reasoning from Sparse Observations},
  author={Ke, Fucai and Cai, Zhixi and Li, Boying and Chen, Long and Lin, Beibei and Wang, Weiqing and Haghighi, Pari Delir and Haffari, Gholamreza and Rezatofighi, Hamid},
  journal={arXiv preprint arXiv:2603.16506},
  year={2026}
}