This week, NVIDIA unveiled a new model of artificial intelligence that has the ability to convert 2D content into 3D content with high accuracy, according to aitnews.
The American tech giant said in a blog post that the new model, called Neuralangelo, can reconstruct 3D using neural networks.
And NVIDIA added that the new artificial intelligence model can convert two-dimensional video clips into detailed three-dimensional structures, and this leads to the creation of virtual replicas and realistic buildings, sculptures and other things in the real world.
Professional content creators can import 3D images generated from 2D content into design applications, and modify them for use in art, video game development, and robotics.
NVIDIA said Neuralangelo’s ability to translate complex material textures, including: ceiling panels, slabs of glass, and smooth marble, from 2D video clips to 3D objects is far superior to previous methods.
And the higher fidelity of 3D reconstructions makes it easier for professional developers and content creators to quickly create usable virtual objects for their projects using smartphone-captured footage.
In a demonstration, NVIDIA researchers showed how the model could recreate famous objects, such as Michelangelo’s Statue of David, or common objects, such as a truck. The Neuralangelo model can also reconstruct the interiors and exteriors of buildings.
Using a 2D video of an object or scene shot from different angles, the model selects several frames that capture different points of view. After positioning the camera for each frame, the AI creates an approximate 3D representation of the scene.
The model then refines the rendering to make details clearer, just as a sculptor carves stone to mimic the texture of a cloth or a human figure, and the end result is a 3D object or large-scale scene that can be used in virtual reality applications or robotics development.
And NVIDIA said that Neuralangelo is one of about 30 projects working on the research department, NVIDIA Research, that will be presented at the Computer Vision and Pattern Recognition Conference, which will be held between June 18 and 22 in Vancouver, Canada.