| --- |
| license: cc-by-nc-sa-4.0 |
| --- |
| |
| # Dynamic Objects Dataset |
|
|
| This dataset is proposed by [NVFi](https://github.com/vLAR-group/NVFi), and used by [FreeGave](https://github.com/vLAR-group/FreeGave) and [TRACE](https://github.com/vLAR-group/TRACE). |
|
|
| ## Structure |
| The structure of the dataset is as: |
| ``` |
| DynObjects |
| | - data |
| | | - fallingball |
| | | | - 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 |
| | | - bat |
| | | - telescope |
| | | - fan |
| | | - whale |
| | | - shark |
| ``` |
|
|
| ## Citation |
| If you find this dataset helpful, please consider citing: |
| ```bibtex |
| @article{li2023nvfi, |
| title={NVFi: Neural Velocity Fields for 3D Physics Learning from Dynamic Videos}, |
| author={Jinxi Li and Ziyang Song and Bo Yang}, |
| year={2023}, |
| journal={NeurIPS} |
| } |
| ``` |
