Common House: Semantic Plan Database for GAN’s in Architecture
Matias del Campo, A. Alfred Taubman College of Architecture + Urban Planning
Collaborators: Justin Johnson (CS), Jessy Grizzle (Robotics), Danish Sayed, Alexandra Carlson
One of the main problems today in working with any form of AI in architecture design is the lack of datasets specifically tailored towards architecture design. The existing datasets that contain buildings and plans were, in most cases, not created by architects and thus lack the level of information and attention to detail that only trained architects and architecture students can provide. This project has embarked on creating large-scale repositories that contain annotated plans and house 3D models that would allow everyone to interrogate design based on high-quality datasets.
In order to achieve the scale of data necessary for neural networks to learn features with the necessary depth to create comprehensive projects, thousands of plans and models are necessary. Engaging a broader global community in this effort will expand diversity and will aid in avoiding the pitfalls of creating heavily biased datasets. The resulting datasets will be free and publicly available, in the spirit of open source and equitable access to data. The process has been streamlined entirely to facilitate participation. In addition, a series of tutorial videos is being developed, so you can learn how to use these datasets in design tasks.
If you are interested in working with this project by annotating a couple of plans or contributing with a couple of models, please fill out the application form HERE. All participants will receive collaborator credit and $4 per completed plan.