Interior Plan Generation for Residential Buildings

1. Configuration Python PyTorch Visual Studio Qt
2. Creating the dataset (1) Download the dataset; (2) Create a directory named dataset under the root directory, and divide the dataset into dataset/train and dataset/val; (3) Run python write_pickle.py, and this should create a new directory, pickle, under the root directory.
3. Training the models We provide four training scripts: train_living.py, train_continue.py, train_location.py andtrain_wall.py that you can find in the folder Living, Continue, Location, and Wall, respectively. The neural networks are described in detail in our paper.
4. Synth floor plans

1. Move the trained models into the folder synth/trained_model; These trained models includes: living_fc1_300.pth, living_resnet34_300.pth continue_fc2_300.pth, continue_resnet34_300.pth location_up1_100.pth, location_resnet34_100.pth wall_up1_100.pth, wall_resnet34_100.pth
2. Now, navigate to /synth, and run python synth.py. The output floor plan is in the folder synth_output.

5. Vectorization To vectorize the floor plans generated by our method, run python vectorization.py. This should create a new directory, synth_vectorization, under the root directory. There is also a GUI tool to visualize the vectorized floor plans.