X
Hover the image you wish to focus. Got it!
Original Location
Beautified Location
Urban Scences
Urban Design Metrics
Change of Urban Elements

facelift

Buildings and neighbourhoods speak. They speak of egalitarianism or elitism, beauty or ugliness, acceptance or arrogance. The aim of Facelift is to celebrate egalitarianism, beauty, and acceptance by beautifying the entire world, one Google Street view at a time.

All of this is done by designing state-of-the-art technologies that make it possible to smarten a street view and read inside the Deep Learning "black box". With further developments of these technologies, we would be more likely to systematically understand and re-create the environments we intuitively love.

"Beauty is nothing other than the promise of happiness"
Stendhal, On Love.

Explore Boston with FACELIFT
No, first tell me more about it.
Discover the street views before-after beautification by clicking on white-circled points. Got it!
Discover where the urban elements are in the city by hovering over or clicking on the labels. got it!

Labels are ordered by beauty, ranging from beautiful to ugly. The bigger, the more frequent.

currently beautiful currently ugly wannabe beautiful, click to compare
The DNA of Beauty

We can compare the changes in Urban Elements (left) and Urban Design Metrics (right) for all beautified locations.
Blue bars indicate an increase in the beautified location, red bars indicate a decrease.

How does it work?
Rating images

Images of urban places are rated by humans in pairs to determine, which one is more beautiful. We then transform these ratings to absolute ranks using TrueSkill.

Training Data

These most beautiful and ugliest images are used as Training Data

Deep Learning

A neural network is then trained on visual cues of beauty and ugliness.

Beautiful and Ugly Combinations

An original (ugly) image is then used as Input for the Beautification Process. The network generates a template using Generative models, maximizing the visual cues for beauty in accordance with the network’s knowledge. Since this template does not (yet) look like an actual place, we search for similar images in our database to find the closest match.

Analysis and Computation

Both, the original and the beautiful images are then analyzed: PlacesNet detects possible Scene types in the image. SegNet shows, what urban elements are visible in the image. Using these insights, we can calculate Urban Design Metrics.

Visualization & Comparison

As a result, we can display the visualizations above

Sagar Jogeklar & Tobias Kauer, Nokia Bell Labs 2017