Unit 3

Reflection, information,
and Machine vision sampling

Jimmy Wei-Chun Cheng

Texture and color of Contemporary Architecture, Neural style transfer.
Texture and color of Seoul, Neural style transfer.


The style of architecture is distinctly connected to art, politics, economy, or religion in any period of history. Any kind of form, color, or texture could serve a cultural meaning in its visual appearance. In other words, the configuration of the building and the material of the skin contains the spirits of time. Based on this concept, buildings transform into objects that represent something more than an inhabitable space. The relationship between space and form is associated with the cultural context of a city. Buildings inside the city have their own roles to sustain the city’s operation. Each type of building congregates to reflect the culture of a city. A city is a composite aggregation of building components. What are the components that make Seoul unique? What are the colors or textures that can represent Seoul?

“Apophenia” by its definition: the tendency to perceive a connection or meaningful pattern between unrelated or random things. Humans tend to recognize things that are related to what they are familiar with. For other people without the same knowledge or experience, the familiarity emotion would not be able to convey. “K-Apophenia” is an act to rethink what can represent Seoul? What are the elements of Seoul? What are the colors of Seoul? What are the forms of Seoul? Can the individual memories of Seoul be shared collectively?


The workshop will kick off with a lecture and discussion of architectural style and elements from historical and contemporary points of view. Students will be required to construct their statements to reflect this topic. In addition, a public library program will be assigned as the main program to execute this workshop. There will be tutorials, lectures, and work sessions each day to develop daily assignments. Any previous knowledge in 3d modeling and visual scripting is not required.

Part 01: Research

Site Survey:

A machine vision and site survey lecture will be conducted on day one. Students can choose a site from one of the selected sites based on their reflection on the topic of style and elements. Ideally, each site reflects different aspects of Seoul. Next, students will spend two days observing, experiencing, and analyzing the context of the site. All the collected images will be used as the dataset for the next step.

Precedent Study:

A series of architectural precedents will be presented chronologically. Students will choose their direction to collect a data set that consists of plans and exterior images. The data set will not be limited to the architectural precedent. It can be any kind of object or artwork which is in the same timeframe. Next, students will build a 2D and 3D Asset library based on the Research materials.

Part 02: Production

Production will be split into four parts. First, students will train a Neural Networks to extract the feature from the precedents’ plan and images. Secondly, using style transfer and AI image classification to initiate the formal massing study. Thirdly, design the detail components which will later be deployed to the massing model as part of the architectural elements on the skin. Finally, using pix2pix to organize the internal program relationship of the library.



Blender plug-in: AI image classifier + Neural Style Transfer + Pix2Pix Gan


Expected outcome

The primary objective of this unit is to see a city in a different way and provide an understanding of machine vision in architectural design. This unit will focus on NN Machine learning, Style Transfer, and 3D modeling.

At the end of this course, students should be able to:

1. To understand how to use a few Machine learning in architectural design.

2. To explore ideas through machine vision sampling.

3. To learn how to create a new environment in Anaconda.

4. A design proposal, with deliverables:

1) Collection of 3D massing model

2) Collection of the sampled imagery dataset

3) 1 Architecture plan, 1 Architecture section

4) 15-second interior rendering or 3 interior rendering

5) 15-second ariel view turntable or 3 exterior rendering