[ad_1]
NVIDIA has launched a project to create realistic photos of fictional characters using artificial intelligence, specifically GAN and style transfer technique.
Everything is done thanks to the automatic identification and separation of the basic attributes of the photo as pose and identity, with more subtle variations like glbades, buttons, hair, marks on the skin and so on. Then, these attributes can be mixed to create different non-existent faces.
The result of these combinations, which you can see in the video below, is is not anything but (19659009)
The video has been posted (unlisted), but this is not a problem, it is very disturbing, especially if we think that None of the faces shown in the video belong to a real person, on the YouTube channel of Tero Karras, one of the newspaper's authors, alongside Samuli Laine and Timo Aila, all employees of NVIDIA Research. Going beyond human faces, they also carried out badessments with other series of photos, showing rooms, cars and cats, which you can also see at the end of the video above and in the picture below. In addition to faces, NVIDIA has generated images generating images of parts, cars and cats using AI. "width =" 680 "height =" 296 "srcset =" https://meiobit.com/wp-content/uploads/2018/12/20181218nvidia_quartos_carros_gatos_ai-680×296.jpg 680w, https://meiobit.com/wp-content /uploads/2018/12/20181218nvidia_quartos_carros_gatos_ai-768×335.jpg 768w, https://meiobit.com/wp-content/uploads/2018/12/20181218nvidia_quartos_carros_gatos_ai.jpg 1000w "sizes =" (max-width: 680px) 100vw, 680px "/>
A few months ago, the same researchers published another project based on a celebrity database, the difference being that the current one is made up of a much larger group of celebrities. Images of faces, with more than 70,000 high-resolution human faces collected on Flickr, of different ages, genders, ethnicities and backgrounds
The other day, NVIDIA released the video of a demonstration of a game containing graphics entirely based on artificial intelligence, also with the help of a GAN. , you can clearly see the path they want to follow with all these imaging and character imaging technologies using AI
Recently, NVIDIA has created a way to fill in deleted areas of photos using of neural networks and artificial intelligence, see this Cardoso post. Another interesting NVIDIA project that became his post a few years ago was a partnership with Adobe, a way to print digitally painted images using 3D brushstrokes.
Talking about brushstrokes, investing in network generation. opponents could be many, let alone those who are responsible for the artwork that was auctioned at an exorbitant price at Christie's this year, a much higher figure than many works of art made by real artists.
To read the printed page The attack network generator architecture generative style-based is available. If you want to access the PDF file directly, click here.
Source link