Facebook working on AI to fix accidental pictures
Generative adversarial networks (GAN) being used to learn about the minute details of the eye
The results are as good as they could get. Most people tricked!
Be it a birthday party, a date or a casual outing with friends, photographs are mandatory. Once in a while it does happen that a great picture is ruined because of the eyes being closed. We are all familiar with re-clicking a picture because of this exact reason. But what if you figure it out later? Neither do we have a time machine nor are we magicians to fix it! There have been a few basic apps that were attempting to tackle this problem but the results haven’t been convincing. But now, Facebook has come up with an “AI fix” to this problem!
Yes! You read it correctly. Facebook has the solution.
The techies at Facebook have adopted a technique called as “in-painting” which basically fills the voids in a picture in a trial and error manner. Generative adversarial networks (GAN) have enabled the researchers to develop a method effective enough to fix the problem.
Pretty impressive, isn’t it?
Facebook used Exemplar Generative Adversarial Network (ExGAN) which compared the data of pictures with eyes open to study the dynamics of the eye and thus, was able to match the pair of eyes with the corresponding face.
Feels like 2050, right?
So how does GAN work?
Let’s find out!
Generative adversarial networks (GANs) are a class of AI algorithms used in unsupervised machine learning.
It is implemented by a set of 2 neural networks working simultaneously, generative and discriminating.
The generative network recognizes the face and learns it shape, color and features of the eye.
The discriminating network repeatedly creates images based on the data received from the generative network.
When the system was tested, most of the audience was tricked into believing that the images were original. The success rate of the system is as high as 95% and with the constant feedback and review process, triumph isn’t far!
So Let’s Summarize,
Although in-painting may sound like a new term, it has been around for a while. Existing platforms like Adobe Photoshop have used such methods regularly for image painting but the fact that this AI specifically targets eye-opening makes it a first of its kind.
Well, this definitely does promise better pictures in the future!