Neuralangelo is the latest technological innovation from NVIDIA. Thanks to this tool, we can quickly obtain a 3D design model from a simple 2D video filmed on a smartphone.
Image scaling, 3D generation… NVIDIA has been offering tools boosted with artificial intelligence for years and has built up considerable expertise in the field. The brand today announces a brand new project, Neuralangelo, an AI model based on neural networks, capable of 3D reconstruction from 2D videos.
From a 2D video to a 3D model
Named after the famous Italian sculptor (among many other activities) Michelangelo, Neuralangelo generates “realistic virtual replicas of buildings, sculptures and other real-world objects”explains NVIDIA, all with complex details and textures: roof shingles, windows, smooth marble…
One of the main interests of Neuralangelo is that it then allows professionals and artists to import the 3D objects thus created into their design applications in order to edit and personalize them.
This new AI-powered tool can save considerable time. All you have to do is take a photo with your smartphone to submit it to Neuralangelo and get a first version of the 3D model. Sectors such as art, game development, robotics and manufacturing will particularly benefit from these features.
For developers of video games, virtual reality, or manufacturers
“The 3D reconstruction capabilities offered by Neuralangelo will be a huge advantage for creators, helping them recreate the real world in the digital world”commented Ming-Yu Liu, Senior Director of Research at NVIDIA. “This tool will allow developers to import detailed objects – whether small statues or massive buildings – into virtual environments for video games or industrial digital twins (note: for the virtual representation of a product) “he added.
NVIDIA says previous 3D scene reconstruction AI models struggled in certain scenarios, such as accurately capturing repetitive texture patterns, consistent colors, and strong color variations.
Neuralangelo selects multiple frames from a video to get a variety of viewpoints and angles, then more accurately calculates the depth, size, and shape of objects. The rendering is then optimized to refine detail and mimic texture, whether it’s a single 3D object or a large-scale scene.
Source : NVIDIA