ComPosition

Spring 2020

Team
Group (other members: RO, SO)

Skills
Rhino, Grasshopper, Python

Course
DEA 3306 - Generative Design Studio

Goal
Create a generative tool that can aid the design process.

Contribution
Development of the Model Screen Capture Script

 

The Script and Its Founding Logic.

 

Static versus Dynamic Compositions.

 
tumblr_nrzflhPuxC1r9nwnbo1_1280.jpg
static structure.png
 
dynamic structure.png
unnamed.jpg
 
 

Static Compositions are comprised of mostly horizontal and vertical lines, and evoke a sense of stability and groundness.

 

Dynamic Compositions are comprised of mostly diagonal lines, and evoke a sense of movement and stability.

 
 

The Script Itself.

Illustrated Grasshopper script of the screen capture system (top) and Illustrated Grasshopper script of the neural network system (bottom)

Illustrated Grasshopper script of the screen capture system (top) and Illustrated Grasshopper script of the neural network system (bottom)

 
 

The objective of this tool is to help a user — designer or not — generate emotionally-stimulating renderings of their projects in an efficient and automated manner. It involves helping the user make informed choices about which compositional angles of their work would be best suited for a desired emotional response, therefore cutting down time from the design process and speeding up demonstrations or iterations of ideas.

 

Capture All Angles of Your Model.

Shrunken isolated diagram of previous screen capture system

Shrunken isolated diagram of previous screen capture system

The first step of the tool involves setting up a point-cloud around a user-chosen geometry. The number of points on this cloud can be chosen by the user, and they determine locations where a camera will be positioned, with all locations being faced to a target point at the center of the geometry. This means that at every point of the cloud, a different angle of the geometry will be taken as a screen capture, and these screen captures are saved to a computer folder of the user’s choice.

 
Screenshot from Location Point 1 facing the Target Geometry

Screenshot from Location Point 1 facing the Target Geometry

Screenshot from Location Point 2 facing the Target Geometry

Screenshot from Location Point 2 facing the Target Geometry

Screenshot from Location Point 3 facing the Target Geometry

Screenshot from Location Point 3 facing the Target Geometry

point cloud with labels.png

 The Neural Network: A 3-step Process.

Shrunken isolated diagram of previous neural network system

Shrunken isolated diagram of previous neural network system

The second step of the process involves passing the user’s folder of screen captures through convolutional neural network. This neural network is trained to detect horizontal/vertical and diagonal lines, which are factors used to make judgments on whether an image is static or dynamic, respectively. The classifications made by this neural network are labeled and passed through to another computer folder of the user’s choice. This process ultimately informs the users of which geometry angles yield more static or dynamic compositions.

Step-by-step flow of the neural network function, from the original data set, to labeling, to static/dynamic classification.

Step-by-step flow of the neural network function, from the original data set, to labeling, to static/dynamic classification.