scintigimcki/DynamicIndoorScenes

Dynamic Indoor Scenes Dataset

This dataset is proposed by NVFi, and used by FreeGave and TRACE.

Structure

The structure of the dataset is as:

DynObjects
| - data
| | - darkroom: data for Gnome House scene
| | | - 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
| | - chessboard: data for Chessboard scene
| | - dining: data for Dining Table scene
| | - factory: data for Factory scene

Citation

If you find this dataset helpful, please consider cite:

@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}
}