Ongoing Project

PlacemakingAI

Harvard Graduate School of Design

This project proposes a novel method for generating and visualizing walkable urban streets supported by Generative Adversarial Networks (or GANs). The presented images synthesize a collection of urban image data classified through metrics of walkability, including green, commercial, and pedestrian streets. A customized Pix2Pix model (Isola et al. 2016) is trained continuously for 3 days to generate new urban images in real-time. An accessible user-interface is used to create place, where images or 3D camera views of streets can be overlaid and synthesized based on these pre-trained models. By converging both the generative image methodology and a user interface PlacemakingAI fosters an urban design process for citizens, designers, and stakeholders alike. 

Project Team: George Guida, Dongyun Kim

 

Harvard Graduate School of Design, with Jose Luis Garcia del Castillo Lopez

CAADRIA 2022 Conference Paper Link

Slide35.PNG
Slide34.PNG

Feedback Process

Encoding Place

Slide36.PNG

Machine Learning Models

UI_Interface.JPG

Real - Time User Interface

Slide38.PNG

Pix2Pix Results

Project Video