The dataset focuses on the building angles, allowing us to see the three-dimensional aspect of the building. By creating our own dataset, and taking quality images from google streetview (780 images), we could start an efficient learning process of the GAN. The 4 images chosen for the interpolation explore the character of an angle.
The image is selected considering different factors. What geometry is the angle ? Is a curved angle considered as Zurichness? Is the angle visible from base to top or is it interrupted by horizontal lines ? Can there be windows in the angle ? Mixing these different types of façade and exploring the latent space give us interesting suggestions.
The image reveals an unfolded façade, by breaking the angle and creating a smoother and curved transition between the two sides of the building. The horizontal façade is accentuated by horizontal lines under the vertical windows. A clear horizontal line that envelops the building separated the last upper floor from the rest of the building. The composition of the ground floor is different than the rest of the building : it is perforated by columns and has a bigger height. The windows of the upper floor are treated in a more dense way.