This AI Can Make A Neighborhood Looks Beautiful In Seconds

Aadhya Khatri - Jan 23, 2020


This AI Can Make A Neighborhood Looks Beautiful In Seconds

Ugly urban areas are an eye sore and British scientists have found a way to make them look more appealing with the help of AI

Ugly urban areas are an eyesore and some British scientists have found a way to make them look more appealing with the help of AI.

Their invention is called FaceLift and what it does is to show before and after pictures of the same neighborhood. One is the current state and the other is the outcome if we follow the beautification methods it proposes.

Computers and AI have become quite skillful in assessing if a certain neighborhood will win the love of humans by analyzing our preferences. Some of the criteria are coziness, the amount of greenery, and visual richness.

AI-urban-areas
Computers and AI have become quite skillful in assessing if a certain neighborhood will win the love of humans by analyzing our preferences

The results have been encouraging and AI has had lots of success in predicting if people will feel safe in a particular place.

Before FaceLift, all machines could do is to give projections of whether people will love a place or not. However, experts at Nokia Bell Labs and King’s College, London think they could do better than that. They envisioned a system that can also propose changes to improve how a neighborhood looks.

According to Sagar Joglekar, the lead author of the project, this challenge has never been tackled before.

What they did to prepare for the AI is to determine the five criteria people use to evaluate how beautiful a place is. They are greenery, walkability, the feeling of openness, memorable landmarks, and visual complexity.

The scientists recruited a number of volunteers to assess 20,000 images and label them ugly or beautiful. Google street view’s resource is also used.

These labeled pictures act as the feeding data for the AI, which works just like human brains when it comes to data processing to make sure that the computers have the same idea as humans in what is considered beautiful. Next, the AI improved the ugly scenes according to the five aforementioned parameters.

AI-urban-planning-generative-adversarial-network
A generative adversarial network comes in to play here. The system is relatively new and has been used to recreate images, including human faces

The AI will add greenery here and there, change the location of the parking lot, and so on. However, as the results look far from realistic, it has to take another step, which is to search an image library to find one that resembles its idea. The AI will also explain why it did what it did to the scene.

Joglekar said that the machines did the job with flying colors. The proposed plans make some neighborhoods twice as walkable and human volunteers often agree with 77.5% of what the computers generate.

However, the system is not without drawbacks. Sometimes it suggests unrealistic alterations like widening the roads in a densely populated area or moving a whole building elsewhere.

These limitations mean that FaceLift cannot replace human planners, at least for now, but it can be a handy tool for the job.

Joglekar said that the team did not expect the AI to rival professional urban planners but the system did have an edge in the speed at which it can generate a more beautiful image for an entire city.

>>> Toyota Is Building A High Tech City Where AI, Robots, And Humans Live Alongside Each Other

Tags

Comments

Sort by Newest | Popular

Next Story