AI & URBANISM
Scanning, classification and generation of urban typologies through deep learning
̌
Deep Urbanism
Scanning, classification and generation of urban typologies through deep learning
SEMANTIC SEGMENTATION
Image Classification at the Pixel Level
Samples of region were annotated to identify buildings that fall within certain typologies.
DATA AUGMENTATION
FINE TUNING THE MODEL
Pixel Classification
Traced Vector Boundaries
Processed Boundaries
EXPERIENCING HIGHER DIMENSIONAL DATA
FEATURE EXTRACTION 1
The first atlas visualization lays out buildings according to orientation and area of scanned boundaries
PLOTTING HISTOGRAMS
Distribution of Boundaries
Orientation by Area
TOWARDS AN ATLAS OF TYPOLOGY
IMAGE TO IMAGE TRANSLATION
Using Generative ADVERSARIAL Networks (GANS)
The following are samples of images generated from a neural network trained on one typology. The neural network successfully generates deep structure layouts for edge cases and everything in between. Furthermore, the networks generates layouts for inputs outside its training set while still maintaining heuristics for the urban typology.
FEATURE EXTRACTION 2
The Second atlas visualization lays out buildings according to aspect ratio and SOLID/VOID RATIO of boundaries
PLOTTING HISTOGRAMS
Distribution of Deep Structures
Aspect Ratio by Solid/Void Ratio
Boundary
Deep Structure
Plan Layout
