Neuralangelo

Neuralangelo is an AI model designed to reconstruct 3D scenes from 2D video data. It targets creative professionals and developers working with art, design, and immersive environments, enabling lifelike replicas of real-world objects and scenes by turning footage into usable 3D assets. The model emphasizes high fidelity textures and detailed geometry to assist in visualization, game development, robotics, and industrial digital twins. It focuses on translating real-world footage into shareable digital representations.

Using Neuralangelo involves selecting frames from a 2D video to capture different viewpoints, determining camera positions, creating a rough 3D representation, and then refining the render to sharpen details. The process mirrors a sculptor starting with rough shape and then chiseling texture and form to closely mimic real material appearance and fabric or figure textures.

The final output is a 3D object or large-scale scene that can be used in virtual reality applications, digital twins or robotics development. It supports recreating objects from everyday items to architectural scale, highlighting high-detail reconstruction capabilities for practical digital fabrication and visualization workflows.

Neuralangelo is introduced in a NVIDIA Research context and is demonstrated as a method to reconstruct 3D scenes from 2D sources, illustrating its potential to translate two-dimensional footage into three-dimensional digital assets. The approach is positioned within ongoing research and its availability on GitHub is noted.

Use cases

• Reconstructs 3D scenes for virtual reality, digital twins, and robotics
• Replicates textures of real-world materials for detailed assets