Publication of "FaceLift: a transparent deep learning framework to beautify urban scenes" in the Royal Society Open Science journal.
FaceLift is a deep learning driven algorithm that goes beyond assessing beauty in urban scenes (streetviews) and recreates the elements that can make any given urban scene more beautiful. It then explains the elements behind the beauty using metrics commonly understood by Urban Planners and Architects.
The study builds a model of urban beauty using 20,000 street view images, and around 1 million votes from crowds. It then uses Generative Adversarial models to mutate an input image into something that maximizes its beauty score. The differences are then explained using 5 urban design metrics: Walkability, Greenery, Complexity, Openness and Landmarks.
Press coverage: