---
task_categories:
- image-to-video
---

# SeeU45 Dataset

This repository contains the SeeU45 dataset introduced in the paper [SeeU: Seeing the Unseen World via 4D Dynamics-aware Generation](https://huggingface.co/papers/2512.03350).

SeeU is a framework that learns continuous 4D dynamics from sparse and monocular 2D frames, enabling the generation of unseen visual content through a 2D $\to$ 4D $\to$ 2D learning process.

- **Project Page:** [https://yuyuanspace.com/SeeU/](https://yuyuanspace.com/SeeU/)
- **GitHub Repository:** [https://github.com/pandayuanyu/SeeU](https://github.com/pandayuanyu/SeeU)

## Sample Usage

### Download Pre-Processed Dataset
You can download the pre-processed SeeU45 Dataset using the `huggingface_hub` CLI as shown in the official documentation:

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

hf download pandaphd/SeeU45_PreProcessed --repo-type dataset --local-dir preproc/SeeU45
```

## 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}
}
```