Houses3K

Next-Best View Policy for 3D Reconstruction to be presented at 2020 ECCV Wokshop.

Peralta, D., Casimiro, J., Nilles, A.M., Aguilar, J.A., Atienza, R., and Cajote, R. "Next-Best View Policy for 3D Reconstruction." European Conference on Computer Vision (ECCV) Workshops, 2020.

Houses3K is a dataset of 3000 textured 3D house models. Houses3K is divided into twelve batches, each containing 50 unique house geometries. For each batch, five different textures were applied forming the sets (A, B, C, D, E).

<img src='imgs/Dataset_Table.png' width="400"/> <img src='imgs/HouseDataset.gif' width="400"/>

Modular approach to creating Houses3K

<img src='imgs/modular.png' width="600"/>

3D Modular Pieces

<img src='imgs/ModularPieces.gif' width="400"/>

Assembly

<img src='imgs/HouseBlockAssembled.gif' width="400"/>

Download

Houses3K Dataset can be downloaded here.

The dataset is in FBX format. You can convert it to other 3D model formats using 3D modelling software like Blender or Maya.

Citation

Houses3K was originally used to train and evaluate Scan-RL. Please cite our paper if you find Houses3K useful to your research.

@article{peralta2020next,
  title={Next-Best View Policy for 3D Reconstruction},
  author={Peralta, Daryl and Casimiro, Joel and Nilles, Aldrin Michael and Aguilar, Justine Aletta and Atienza, Rowel and Cajote, Rhandley},
  journal={arXiv preprint arXiv:2008.12664},
  year={2020}
}

License