FACEBOOK IS MESSING AROUND with artificial intelligence (AI) again, this time using it to open people's eye in selfies where they've blinked.
Using a machine learning method called generative adversarial network, a technique which has been used to generate images using AI tech such as superimposing famous celebrities faces over those of pornstars, Facebook's AI can retouch closed eyes in photos.
The tech learns what a person normally looks like by being trained on images where their eyes are open, thereby building up a picture of the characteristics of a person's eyes, such as colour and shape.
Once trained, the AI can then use that information to carry out a process dubbed "eye in-painting", essentially filling in the blinking eyes with fully open peepers.
The tech currently exists as research by Facebook's boffins, and has a few shortcomings, such as struggling to with people who wear glasses or have taken selfies at difficult angles. So it's not likely to be rolled out into Facebook's main social network any time soon.
But the tech demonstrates the scope for AI to be used in manipulating images and fixing human mess ups. It also offers an alternative in AI image manipulation to deep learning neural networks, which tend not to preserve the true identity of a person, which a generative adversarial network can do.
This also showcases the potential for AI to produce more fake images or manipulate photos to make every selfie or random snap look perfect. In turn, this raises questions around authenticity and how real and true people's social media personas are to their real-world selves.
But then we have phones with 'beauty modes' and the ability to finely retouch pictures without delving into Photoshop, so one could argue that a level of fakery is already happening and Facebook's AI is just an evolution of that. µ
It's an onomatopoeic week for Google
Hope that free lunch was delicious
It's like Bixby being terrible never happened
Notch to be outdone