June 2 (Portaltic/EP) –
nvidia has unveiled a new artificial intelligence (AI) model that can convert two-dimensional (2D) video into detailed three-dimensional (3D) structures and generate highly realistic virtual replicas of buildings and sculptures.
Is about Neurangelo, that has the ability to translate the textures of different materials (such as tiles, glass or marble) with high fidelity, which can help art, video game and robotics designers to recreate real-life objects.
The team in charge of developing this AI model, NvidiaResearch, has explained that to use it, you only need a two-dimensional image or a video captured with a smartphone. Once you collect the content information, presents a three-dimensional replica of that objectalthough it can also rebuild interiors and exteriors of buildings.
The company has recalled that previous AI models intended for the reconstruction of 3D scenes “have had problems accurately capturing repetitive texture patterns, homogeneous colors and strong color variations”, as it has pointed out in a statement.
To improve results, Neurangelo adopts instant neural graphics thanks to Nvidia Instant NeRF, which helps capture the finest details of the objects or spaces you want to replicate in 3D.
First of all, use a 2D video of an object or scene shot from multiple angles. After choosing a series of frames that capture different viewpoints, the AI ​​determines the camera position of each frame and creates an approximate 3D representation of that subject or location.
The model then optimizes the rendering to fine-tune details and mimic textures. This, by the way, can be used in virtual reality applications (RV), those dedicated to the development of robotics and digital twins.
