NVIDIA Get3D

What is NVIDIA Get3D?

NVIDIA’s GET3D is a generative model that creates detailed, textured 3D meshes directly from 2D images. It was developed in collaboration between researchers at NVIDIA, the University of Toronto, and the Vector Institute, and was presented at NeurIPS 2022. Unlike earlier 3D generative models, GET3D produces high-quality, textured meshes that work smoothly with standard 3D rendering tools.

A standout feature is text-guided shape generation: you can type a description and the model uses it to help create the 3D shape you’re looking for. The model is also trainable end-to-end, with a clear separation between geometry and texture, supporting both efficiency and creative control.

Who created NVIDIA Get3D?

GET3D was created by a team of researchers from NVIDIA, the University of Toronto, and the Vector Institute, who presented their work at NeurIPS 2022. Contributors include Jun Gao, Tianchang Shen, Zian Wang, Wenzheng Chen, Kangxue Yin, Daiqing Li, Or Litany, Zan Gojcic, and Sanja Fidler.

What is NVIDIA Get3D used for?

• Creating high-quality 3D assets: Generates textured 3D shapes with intricate details from 2D images.
• Geometry/texture disentanglement: Keeps geometry and texture separate, giving more creative freedom.
• Text-guided shape generation: Create shapes by typing text prompts.
• End-to-end training: Uses adversarial losses and differentiable rendering for efficient training.
• Unsupervised material generation: Produces materials and view-dependent lighting effects without special supervision.
• Diverse shapes: Creates shapes with different topologies, high-quality geometry, and textures.
• Smooth transitions: Interpolate between latent codes for smooth transitions between shapes.
• Local variations: Generate similar shapes with slight local differences.

Who is NVIDIA Get3D for?

GET3D is particularly useful for professionals in:

• The gaming industry
• The film industry
• The virtual reality industry

How to use NVIDIA Get3D?

To use GET3D, start by accessing the generative model itself, which is designed to create high-quality 3D textured meshes directly from 2D images. The underlying approach draws on advancements in differentiable rendering, surface modeling, and generative adversarial networks to produce complex 3D shapes. For full details and access, visit the official GET3D research page.