---
license: cc-by-nc-sa-4.0
---

# FreeGave-GoPro Dataset

This dataset is proposed by [FreeGave](https://github.com/vLAR-group/FreeGave).

## Structure
The structure of the dataset is as:
```
DynObjects
| - data
| | - pen1: 
| | | - train: serves as training data
| | | - val: used for evaluating novel view interpolation
| | | - test: used for evaluating future extrapolation
| | | - transforms_train.json: camera poses and other meta informations for training set
| | | - transforms_val.json: camera poses and other meta informations for novel view interpolation task
| | | - transforms_test.json: camera poses and other meta informations for future extrapolation task
| | | - points3d.ply: randomly initialized points for 3D Gaussians
| | - pen1
| | - box
| | - hammer
| | - collision
| | - mat
```

## Citation
If you find this dataset helpful, please consider cite:
```bibtex
@article{li2025freegave,
  title={FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity}, 
  author={Jinxi Li and Ziyang Song and Siyuan Zhou and Bo Yang},
  year={2025},
  journal={CVPR}
}
```