
Architectural Synthesization
Machine Learning Applications
Course: J-term Exquisite Corpse 2.0 - Machine Learning Applications in Architectural Design
Instructor: Dongyun Kim, George Guida
Date: Jan., 2022
The applications of machine learning to architectural design will begin with an understanding of the benefits and limitations of this technology including bias, intelligence, and creativity. This will then be followed by a series of hands-on workshops covering 2D Style Transfer to 3D object manipulation.
​
This course will be divided into three parts: Dataset collection, GAN Training, and 3D object manipulations. Dataset collection will equip students with an understanding of the emerging agency of designers and the implicit bias ingrained within these. The training of state-of-the-art machine learning models and their manipulation into new 3D forms will be used to challenge an emerging homogeneity in architectural design. These will additionally shed light on the emerging role of the designer and how images can be synthesized into 3D ‘exquisite corpses’.
KEYWORDS:
Machine Learning, Architectural Design, Style Transfer, GAN, Object Manipulation, Image Synthesization
OBJECTIVES
​The tangible skills offered in this workshop will cover the scraping of online datasets, using Python with libraries such as BeautifulSoup, the training of GAN (Generative Adversarial Network) models such as Style Transfer and StyleGAN, and their manipulation through grasshopper workflows. Students can openly extend project topics across architectural, landscape, and urban design interests.
Day 1
Theme: Introduction to AI & Architecture 01
Skillset: RunwayML, ML Workflows, Dataset Collection
Dataset Processing
Runway ML Workflow
- Runway Interface

- Pix2Pix Model in Runway

- Pose Net Model in Runway

- Dataset View in Rhino


Day 2
Theme: Introduction to AI & Architecture 02
Skillset: Runway ML, 2D StyleGAN training, 2.5D Applications with Grasshopper
GAN Training in Pytorch
2D StyleGAN Training - Villa Savoye Elevation



2.5D Application with Grasshopper - Latent Walk Mapping



Day 3
Theme: Advanced 3D Applications with Grasshopper
2.5D Stacking Workflow


Connection to Runway ML

Day 4
Theme: Concluding Remarks














