NVIDIA introduces fVDB, a deep-learning framework for creating AI-ready virtual representations of the world

From NVIDIA: 2024-07-29 16:30:47

NVIDIA unveiled fVDB, a new deep-learning framework at SIGGRAPH for creating AI-ready virtual representations of the real world. It’s based on OpenVDB and focuses on generative physical AI like autonomous vehicles and robots, requiring spatial intelligence for operating in 3D space.

fVDB translates raw data from various sources like NeRFs and lidar into massive, real-time rendered environments to train AI. With 4x larger spatial scale, 3.5x faster processing, and 10x more operators, it enables industries to enhance spatial intelligence for physical AI on a larger scale and in higher resolution.

fVDB will soon be available as NVIDIA NIM inference microservices, allowing businesses to integrate it into OpenUSD workflows for generating AI-ready environments in NVIDIA Omniverse. OpenVDB, known for its use in visual effects, has evolved to benefit industrial and scientific applications such as industrial design and robotics.

NVIDIA has been innovating with OpenVDB over the years, from introducing NanoVDB for GPU support to NeuralVDB for machine learning compression. With fVDB’s AI operators built on NanoVDB, spatial intelligence can now be unlocked at reality scale, offering increased capabilities for real-time simulations and rendering. Apply for early access to the fVDB PyTorch extension and find it on the OpenVDB GitHub repository.

Delve deeper into fVDB through a technical blog and watch NVIDIA’s CEO discuss the impact of accelerated computing and generative AI on industries at SIGGRAPH.



Read more at NVIDIA: NVIDIA Introduces fVDB to Build Bigger Digital Models of the World