Datasetpublic

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 | | | -… See the full description on the dataset page: https://huggingface.co/datasets/scintigimcki/DynamicIndoorScenes.

sourceHugging Facecc-by-nc-sa-4.0updated 1y agoView on Hugging Face

1likes66downloads

Dataset Card

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:

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}
}