GET3D | Nvidia

GET3D introduces a breakthrough approach to 3D content creation with its generative model capable of producing high-quality, textured 3D shapes directly from 2D images. Developed by the Toronto AI Lab and presented at NeurIPS 2022, this innovative technology addresses the growing demand for varied, detailed, and ready-to-use 3D assets in industries creating massive virtual worlds. By leveraging advancements in differentiable rendering and surface modeling, along with generative adversarial networks, GET3D can produce meshes featuring complex topologies and rich textures. The technology's end-to-end trainable model, sophisticated geometry-texture disentanglement, and the ability to guide shape generation through textual prompts demonstrate GET3D's commitment to fostering creativity and efficiency in 3D modeling. The model's versatility and its potential to revolutionize industries like gaming, film, and virtual reality make it an exciting development in AI-driven content creation.

Key Features

• High-Quality 3D Assets: Generates 3D textured shapes with intricate details directly from 2D images.

• Advanced Disentanglement: Achieves clear separation between geometry and texture allowing creative flexibility.

• Text-Guided Shape Generation: Offers capability to create shapes based on textual prompts enhancing user interactivity.

• End-to-End Trainable Model: Utilizes adversarial losses and differentiable rendering for an efficient training process.

• Unsupervised Material Generation: Produces materials and view-dependent lighting effects without supervision.

FAQ

What is GET3D?

GET3D is a generative model that creates high-quality 3D textured meshes from 2D images by using advancements in differentiable surface modeling, differentiable rendering, and 2D Generative Adversarial Networks.

What kind of 3D textured meshes can GET3D generate?

GET3D can generate textures and geometries such as cars, chairs, animals, motorbikes, human characters, and buildings with complex topology and rich details.

Who developed GET3D and where was it presented?

GET3D is developed by the collaborative effort of researchers from NVIDIA, the University of Toronto, and the Vector Institute, and was featured at NeurIPS 2022.

How is GET3D different from previous 3D generative models?

GET3D differentiates itself from prior works by its ability to produce explicitly textured 3D meshes that are of high quality and can directly be consumed by standard 3D rendering engines.

Does GET3D support text-guided shape generation?

Yes, GET3D's model allows for text-guided generation where users can input text prompts to guide the creation of 3D shapes, enhancing the generative process with textual input for more meaningful results.