Using deep learning to generate design spaces for architecture

Abstract

We present an early prototype of a design system that uses Deep Learning methodology—Conditional Variational Autoencoders (CVAE)—to arrive at custom design spaces that can be interactively explored using semantic labels. Our work is closely tied to principles of parametric design. We use parametric models to create the dataset needed to train the neural network, thus tackling the problem of lacking 3D datasets needed for deep learning. We propose that the CVAE functions as a parametric tool in itself: The solution space is larger and more diverse than the combined solution spaces of all parametric models used for training. We showcase multiple methods on how this solution space can be navigated and explored, supporting explorations such as object morphing, object addition, and rudimentary 3D style transfer. As a test case, we implemented some examples of the geometric taxonomy of “Operative Design” by Di Mari and Yoo.

Generating Conceptual Architectural 3D Geometries with Denoising Diffusion Models

“…The final dataset is then more than a combination of all the parametric models' outputs as it also contains hybrids of those initial models. We have shown that a generative neural network is able to create a diverse solution-space that maps out all the parametric models but also their interpolation and combinations (Sebestyen et al, 2023). Furthermore, we showed that this new solution-space can be linked to descriptive keywords, which lets the user intuitively navigate it and thus generate novel geometries through semantic inputs.…”

“…2019) which contradicts one of the main principles of deep learning, which is to acquire large datasets in order to achieve the most accurate results. In previous projects we have solved this problem by creating large datasets through parametric scripts (Sebestyen et al, 2023). This is achieved by randomizing parameters' inputs and thus creating a single output instance.…”