README.md

metadata

license: apache-2.0
size_categories:
  - 10K<n<100K
tags:
  - 3D
  - 360Video
  - 360Image
  - Omnidirectional
  - '360'

VegQi/Move360

A high-quality 360° video dataset at 7680×3840 resolution and 30 fps.

Data Organization and Schema

Each row corresponds to a single image frame (or a single image instance). The dataset is represented as a tabular structure with the following fields:

• scene_id: Identifier of the scene/clip segment (e.g., "0001").
• frame_id: Identifier of the frame, derived from the filename stem without the extension (e.g., "00000001").
• relpath: Original relative path in the raw directory structure (e.g., "0001/00000001.jpg").
• image: The image content stored as a datasets.Image feature; it can be decoded directly into a PIL image object when accessed.

---

Quickstart (Python / datasets)

Installation

pip install -U datasets pillow

Load the dataset

from datasets import load_dataset

ds = load_dataset("VegQi/Move360", split="train")
print(ds)
print(ds.features)

Access samples and decode images

sample = ds[0]

scene_id = sample["scene_id"]
frame_id = sample["frame_id"]
relpath  = sample["relpath"]
img      = sample["image"]  # PIL.Image.Image

print(scene_id, frame_id, relpath, img.size)

Citation

If you find the dataset helpful in your research or work, please cite the following paper:

@article{li2025omnidrag,
  title={OmniDrag: Enabling Motion Control for Omnidirectional Image-to-Video Generation},
  author={Li, Weiqi and Zhao, Shijie and Mou, Chong and Sheng, Xuhan and Zhang, Zhenyu and Wang, Qian and Li, Junlin and Zhang, Li and Zhang, Jian},
  journal={International Journal of Computer Vision (IJCV)},
  year={2025}
}