Building-GAN

Code and instructions for our paper:

Building-GAN: Graph-Conditioned Architectural Volumetric Design Generation, ICCV 2021.

Volumetric Design Process

Data

• Download the dataset here.
• Put the subfolders and files in raw-data under the folder 6types-raw_data.
• Run Data/process_data.py to process the raw data.

For the detail about how the raw data are processed, please refer the Data/process_data.py.

In the dataset, each volumetric design comprises three json files:

• Global Graph: contains the FAR, program ratios, and the associated rooms for each program type.
• Local Graph: contains the bubble diagram--the type and size of each room and the connectivity between rooms
• Voxel: contains the voxel graph

Running pretrained models

For running a pre-trained model, please follow the steps below:

• The pre-trained model is located at runs/iccv2021/checkpoints/
• Run python inference.py
• Check out the results in the inference/{model}/{epch_current_time}/output folder.
• Check out the variation results from the same program graph in the inference/{model}/{epch_current_time}/var_output* folders.

Training models

For training a model from scratch, please follow the steps below:

• Follow the steps in Data section.
• run python train.py . Customized arguments can be set according to train_args.py.
• Check out output and checkpoints folders for intermediate outputs and checkpoints, respectively. They are under the runs/run_id/ where run_id is the serial number of the experiment.

Requirements

• PyTorch >= 1.7.0
• PyTorch Geometric 1.6.2

Citation

@article{chang2021building,
  title={Building-GAN: Graph-Conditioned Architectural Volumetric Design Generation},
  author={Chang, Kai-Hung and Cheng, Chin-Yi and Luo, Jieliang and Murata, Shingo and Nourbakhsh, Mehdi and Tsuji, Yoshito},
  booktitle={International Conference on Computer Vision},
  year={2021}
}

Contact

Unfortunately this repo is no longer actively maintained. If you have any question, feel free to contact Chin-Yi Cheng @chinyich or Kai-Hung Chang @kaihungc1993

License

Shield:

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.