The code in this public repository is select code from the Master's Thesis '3D Generative Adversarial Networks to Autonomously Generate Building Geometry' by Lisa-Marie Mueller, completed at TU Delft in June 2023. Once the University posts the thesis and makes it public, the entire thesis can be found on TU Delft's education repository at https://repository.tudelft.nl/islandora/search/?collection=education

The base architectures tested.

Architecture versions A through H as described in depth in the thesis report.

Architecture versions J through W as described in depth in the thesis report.

Architecture versions 12 through 17 as described in depth in the thesis report.

Architectures that were tested for different input types as described in depth in the thesis report.

After installing a Conda environment using the provided environment file, it is possible to run the Jupyter Notebooks which demonstrate some specific parts of the project and make these easier to use and understand. The pre-processing Jupyter notebook demonstrates the steps that were implemented to clean the point cloud models so that they can be used for training GANs. The generate Jupyter notebooks allow users to generate new building forms with the trained models using architecture 16R and 17R, the best performing architectures.

The run files allow you to train your own model using architecture 16R and 17R, the best performing architectures. Please ensure the proper dependencies are installed by using the provided environment file.