Resumen de Flexibility of cGANs to Encode Architecture Isometrics in Colour Patterns

Diego Navarro Mateu, Oriol Carrasco Carrasco, Pedro Cortés Nieves

• Architectural practice as an abstract concept is based on developing projects composed different levels of abstraction. From the conceptualisation of spaces to the expression of geometry, there is hierarchical information that can be abstracted as a pattern. The present research raises the possibility of compressing and encoding this information into colour grids that can be interpreted by an artificial intelligence (generative adversarial network), so that it can represent an architectural object. The development of machine learning allows us to introduce new workflows for both design and analysis. Through image association, this research generates isometric perspectives that represent relatively complex architectures from a simplified chromatic code. In this research, the training of two neural networks tests the possibility of applying indirect coding, as well as the networks’ response and flexibility to external elements outside the initial databases.