FIFA 18’s player faces may look realistic, but for one gamer, they’re still some way off.
Chintan Trivedi took it upon himself to make the in-game faces more lifelike – with incredible results.
He used Deep Neural Networks that he ‘trained’ to replace the existing faces in the game – much like the way Deep Fakes works.
So, how did he do it?
Firstly, to gather the data required for the Deep Fakes algorithm, Trivedi recorded Cristiano Ronaldo’s face from the instant replay option in the game.
He then downloaded a load of images of Ronaldo from Google – making sure to get the star from different angles.
Now, this is where it gets complicated.
Deep Fakes works by neural networks called autoencoders. Trivedi used two of them.
One learns to recreate Ronaldo’s face from FIFA 18’s graphics. The other learns how to recreate the Portuguese star’s face from actual pictures.
In other words, they build an understanding of what Ronaldo looks like in the game and in real life.
The algorithm then swaps the faces around – feeding the real Ronaldo face on to the FIFA input.
The result? Well, see for yourself.
This technology is proof that photorealistic player faces may only be a year or two away. And as you can see, the difference is striking.
Let’s hope EA Sports is taking notice.
You can watch the whole video on Trivedia’s DeepGamingAI channel below: