[ad_1]
Samsung Artificial Intelligence researchers have developed a way to animate classic portraits and bring them to life.
In a new article, researchers from the Samsung Artificial Intelligence Center in Moscow, as well as from the Skolkovo Institute of Science and Technology, explain how they built a human face modeling system using the fewer images possible.
The results are based on existing deepfake techniques, which are used to graft a person 's face onto a video of someone else.
Read more: The artificial intelligence technology behind the scary celebrity deepfakes is used to create completely fictitious faces, chats and Airbnb listings
The Deepfake software usually relies on capturing a lot of data on a person's face, usually using video clips. The new technique focuses on the use of some photos, or even a single still image.
They did this by feeding the software with an extensive set of YouTube celebrity video data on the conversation lead, to train them to identify meaningful "landmarks" on human faces.
It works in conjunction with a Generative Gator Network (GAN), which opposes two algorithms against each other – one generating fake images and the other trying to To catch by spotting which images are wrong.
Here it is applied to the photos of Marilyn Monroe and Salvador Dalí:
The technology is clever enough to identify human faces and can even be applied to old paintings, such as Mona Lisa:
The researchers noted that the use of a single image makes the software less efficient. You can see that the animated versions all have a distinct "personality", that they derive from the people on which they are modeled.
You can watch a full explanation of the results here:
[ad_2]
Source link