---
license: cc-by-nc-nd-4.0
task_categories:
- image-to-video
---

# SeeU45 Dataset

[**Project Page**](https://yuyuan-space.github.io/SeeU/) | [**Paper**](https://huggingface.co/papers/2512.03350) | [**Code**](https://github.com/pandayuanyu/SeeU)

This is the official dataset for **SeeU: Seeing the Unseen World via 4D Dynamics-aware Generation**, a framework that learns continuous 4D dynamics to generate unseen visual content from sparse monocular frames.

## Sample Usage

You can download the dataset using the `huggingface_hub` CLI:

```bash
pip install "huggingface_hub[hf_transfer]"

hf download pandaphd/SeeU45 --repo-type dataset --local-dir SeeU45
```

## Dataset Structure

- `SeeU45_train/`  
  Provides the 2D frames used as model inputs during training.

- `SeeU45_GT/`  
  Contains the full scene ground-truth frame sequences.

- `SeeU45_GT/sample_frame.txt`  
  Specifies which frames are sampled for training. It also records:
  - the original scene/video folder names,
  - sampled frame indices,
  - total frame counts (GT length),
  - and frame numbers corresponding to `SeeU45_train` inputs.

## Data Sources and Licensing

SeeU45 is constructed from a combination of our own captured scenes and publicly available datasets.  
Due to licensing restrictions on some third-party datasets, we are not allowed to re-distribute certain videos.  
As a result, the public release of SeeU45 **omits 5 scenes** that appear in our paper.

The publicly released scenes in SeeU45 are derived from the following sources (in compliance with their respective licenses):

- [TAP-Vid](https://tapvid.github.io/#:~:text=The%20annotations%20of%20TAP-Vid%2C%20as%20well%20as%20the,their%20creators%3B%20see%20the%20DAVIS%20dataset%20for%20details.)
- [AgiBot](https://huggingface.co/datasets/agibot-world/AgiBotWorld-Alpha)
- [I2-2000FPS](https://chennuriprateek.github.io/Quanta_Video_Restoration-QUIVER-/)

All original copyrights of these source datasets are retained by their respective authors.

## License

The **SeeU45 dataset** (including our processed frames, splits, and annotations) is released under **CC BY-NC-ND 4.0**. 

- ✅ You may use SeeU45 for **non-commercial academic research**.  
- ✅ You must credit the SeeU paper and dataset when using it.  
- ❌ You may not use SeeU45 for commercial purposes.

## Citation

```bibtex
@article{Yuan_2025_SeeU,
  title={{SeeU}: Seeing the Unseen World via 4D Dynamics-aware Generation},
  author={Yuan, Yu and Wickremasinghe, Tharindu and Nadir, Zeeshan and Wang, Xijun and Chi, Yiheng and Chan, Stanley H.},
  journal={arXiv preprint arXiv: 2512.03350},
  year={2025}
}
```