NVIDIA Showcases AI Advances in Generative Models, Avatars, and Robotics at NeurIPS 2022

NVIDIA researchers are presenting advancements in generative AI, robotics, and more at NeurIPS 2022. This includes new techniques for text-to-image generation, creating 3D avatars from photos, and building more versatile robots.

Multiple projects focus on improving text-to-image diffusion models, like generating better images from rare or uncommon concepts, completing 3D models from incomplete data, and linking modifiers to main entities.

For AI avatars, new methods efficiently create detailed 3D heads from portraits and enable personalized text-to-speech without recordings.

In robotics, new techniques help reinforcement learning algorithms adapt to diverse tasks and let robots better manipulate unfamiliar objects.

NVIDIA researchers are also applying AI to accelerate simulations in physics, climate science, and healthcare.