A recent research paper claims that researchers at Samsung’s AI Research (Moscow) in association with Skolkovo Institute of Science and Technology have developed software that can create 3D talking avatars from a single still image. Although training it through multiple images will result in higher realism.
Previously, technology was only capable of generating photo-realistic portraits but this whole new approach called “single-shot learning” is can produce moving, talking heads, though not perfect but highly realistic.
What makes this approach different is rather than needing a lot of data (like video & photos) to analyze it needs just one image of a person’s face and then you can see the same face making a range of expressions and actually moving!
Samsung said that the model creates three neural networks during the learning process. First, it creates an embedded network that links frames related to face landmarks with vectors. Then using that data, the system creates a generator network that maps landmarks into the synthesized videos. Finally, the discriminator network assesses the realism and pose of generated frames.
The paper says such ability has practical applications for telepresence, including video conferencing and multi-player games, as well as the special effects industry. However, there’s a huge possibility that the same model can be used to create deep fakes like we’ve seen in the past how the technology has been misused to create celebrity fakes.