Towards Hallucinating Machines

Towards Hallucinating Machines – Designing with Computational Vision

Authors

Matias del Campo, Sandra Manninger, Alexandra Carlson

Keywords

Artificial Intelligence, Design Agency, Neural Networks, Machine Learning, Machine Vision

Abstract

How many thousand images does it take an architect to learn what Gothic is, or Baroque or Modern? How many more to differentiate between good and bad architectural solutions? This article strives to de-mystify the nature of design choice in architecture by interrogating the underlying processes of Neural Networks and thus the extent of their ability to inform architectural design. The presented approach strives to explore the design problem not only through the lens of expediency but also by considering the cultural transformation that comes along with the possibilities of a technology that profoundly asks about the nature of agency in a posthuman environment.

Figure 1: In this example, a neural network is trained to dream or ‘hallucinate’ the features of arches on a given depth rendering to impose novel geometric structure into the input, and then to perform a style transfer between a rendered texture image from a student’s 3D model upon the result of the dreaming process. Works by Hannah Daugherty, Mariana Moreira de Carvalho, Imman Suleiman.

Full Text

Leave a Reply

%d bloggers like this: