SEAN WALLISH

GAN Hadid was my first experiment with DCGAN. I started with the goal of seeing if the a Generative Adversarial Neural Network could carrying on the work of an Zaha Hadid by suggesting new ideas based on the pixel relationships found in images of the previous projects that came from ZHA. A dataset was created by downloading photos and renders of  ZHA projects as well as taking the visualization videos from their youtube account and turning the individual frames in to stills, as a way to increase the numbers of the dataset. 

 

As seen below working at 178x178 pixels returns results that at the full scale of a building suggest less detail then would be required for a more realistic building to emerge from the details. The products of the algorithm do suggest the beginning of building shapes though, and certain aspect of the previous projects emerge in the result. If the viewer has a familiarity with Zaha Hadid's work they will be able to see elements in the results that are referential to her other projects.