OmniDrag

February 13, 2026 · View on GitHub

[IJCV 2025] OmniDrag: Enabling Motion Control for Omnidirectional Image-to-Video Generation

arXiv Project Page visitors Move360 dataset

Weiqi Li, Shijie Zhao, Chong Mou, Xuhan Sheng, Zhenyu Zhang, Qian Wang, Junlin Li, Li Zhang and Jian Zhang

School of Electronic and Computer Engineering, Peking University

🚩 Updates

  • 2026.02.13 Released the Move360 dataset!
  • 2025.11.01 OmniDrag has been accepted by IJCV!
  • 2024.12.12 Released the OmniDrag paper.

🔥 Introduction

Omnidirectional videos generated by proposed OmniDrag. It enables drag-style synthesis from a reference omnidirectional image and user-specified points, providing both scene-level (top) and object-level (bottom) accurate, high-quality controllable generation.

✏️ To Do List

  • Release the training code.
  • Release the inference code and weights.
  • Release the Move360 dataset.
  • Release the paper.

Citation

If you find the code 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}
}