Facebook, the largest and most popular social media platform in the world, has adopted the latest technologies to serve its users with fresh updates. One of the company’s groundbreaking inventions is its AI image recognition technology.
Andrea Vedaldi, one of the AI experts on Facebook when it comes to machine learning and computer vision, said that its new image recognition could transform 2D images into 3D in real-time. Meaning, the algorithms of Facebook’s AI can watch a regular video and convert it into an entirely 3D scene frame by frame as it is playing.
Facebook used the blockbuster action movie The Matrix to demonstrate its AI image recognition technology. On the screen, the scene is playing out as normal. However, while Morpheus and the background are completely the same, Keanu Reeves’s 2D footage has been transformed into a 3D model. The said footage is already pre-rendered. But as mentioned, Facebook Followers can do such transformations in real-time.
Every Follower on Facebook Likes That!
Besides, The Matrix is an awe-inspiring representation of turning 2D images into 3D. The reason is that it has exquisitely choreographed Kung Fu maneuvers that even a human would find difficult to process.
Veldaldi told Digital Trends, an online news outlet, that the said movie is very challenging to process because it shows acrobatic poses. However, she also noted that it is not what people would typically see in a user application. Facebook Followers just used the movie for fun to demonstrate what the system can do.
The algorithms of Facebook are considerably more adept. In addition to that, they are getting way better all the time.
Getting Data from Images
True enough, the new image recognition technology is strange research for Facebook to focus on. It would be more understandable if the company focused on better news feed algorithms and new ways of suggesting content users might be interested in. But turning 2D into 3D images is certainly not something you would expect from a social media giant. However, the research is already in the process, even if there is no immediate plan to turn it into a useful feature on the platform.
For the seven years that passed, Facebook has been continuously working to establish a leading presence in artificial intelligence. In 2013, Yann LeCun, one of the foremost authorities of deep learning in the world, took a job on Facebook. His job is to do AI in a way that would be close to impossible in 99 percent of the world’s AI labs. Since then, the company has expanded FAIR or Facebook AI Research all over the world.
Today, Facebook dedicated 300 full-time scientists and engineers to its goal to come up with the future’s cool artificial intelligence tech. The company has FAIR offices in Pittsburgh, Seattle, New York, Menlo Park, Boston, Montreal, London, Paris, and Tel Aviv, Israel. The said offices are all staffed by several top researchers in the field.
Facebook’s Image Recognition Has Gotten More Sophisticated
The big focus of Facebook is to figure out how it could understand photos better. Since 2017, the social media giant has used artificial neural networks in order to auto-tag people in photos without the users having to manually label them. Since then, Facebook’s image recognition technology became increasingly sophisticated.
But ironically, this was recently highlighted to users when the social media giant faced some problems. A temporary outage in July 2019 prevented many photos from showing up on Facebook. Instead, the images were replaced by borked image frames that have tags generated by machine learnings. Those tags described what the social media giant’s AI thought was in the picture. For instance, an image might contain a tag that reads, “Image may contain: tree, sky, outdoor, nature, cat, people standing.”
Going back to The Matrix, Vedaldi said that the social media giant aims to develop an AI image recognition technology to understand everything in 3D at a single glance. The company also likes to get AI to understand the world, similar to how we do.
How Did They Know We’d Like That So Much?
What Vedaldi said means showing it an image of a plane and having the AI recognize that image as a plane. It must also understand its shape in 3D space, as well a predict how it will move. The same thing goes for other images, such as a car, a bird, or a woman doing yoga.
Moreover, the rationale of the new image recognition technology for Facebook totally makes sense. To be clear, this demo will not appear in a Facebook feature in the coming days. However, the company’s interest is to train an AI to better understand the world through the images that it sees.
In the entire lifespan of Facebook, more than 250 billion images have already been uploaded to the social media platform. This means that every single day, approximately 350 million images are uploaded.
Will Facebook Likes Translate to Instagram Likes?
Also, Facebook is the owner of Instagram, a popular photo-sharing and social media platform. It already has had more than 40 billion photos and videos uploaded ever since its inception. Moreover, some 95 million media are added every single day.
According to Facebook’s mission statement, understanding what goes on in images is extremely important. This is because images are one of the primary ways that people communicate in social media. Being able to interact with and understand images on a three-dimensional plane will also make it possible for Facebook to thrive in new technologies such as augmented reality. Imagine an AR app that could turn your old 2D Facebook photos that you uploaded years ago into 3D images that you can explore in augmented reality. Facebook is not saying that it will develop such a thing. However, the technologies of today are no doubt present to make it, as well as other advanced features possible.