Office Floor Plans Generation Based on Generative Adversarial Network

Zhequan Zhang, MBS '21

ARCH 692B: Building Science Thesis

Instructor: Karen Kensek

Creating a new way of architecture drawing based on machine learning to save architecture or BIM modelers’ time: Based on the BIM data set of an architecture firm, Dynamo (a visual programming language in the Revit) was used to obtain a training set and a test set — two groups of the paired pictures. one type of the GAN called pix2pixHD were trained based on the training set, which is a process of learning the relationship between two paired image. A floor plan with all facilities locations would be predicted by given a certain shape of the room. Then the predicted result is reversed and input back to the rooms with facilities in Revit.