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