NVIDIA fVDB

NVIDIA developed an open-source deep learning framework for sparse, large-scale, high-performance spatial intelligence

It unites fundamental AI operators using VDB. Generative physical AI with spatial intelligence relies on fVDB

fVDB AI operators built NanoVDB, which accelerates OpenVDB GPU. The framework supports sparse convolution and real-time optimised ray tracing

It works with the Kaolin Library for 3D deep learning and Warp for Pythonic spatial computation

In fVDB, differentiable operators like convolution, pooling, attention, and meshing are designed for high-performance 3D deep learning

NVIDIA NIM inference microservices let developers integrate fVDB core architecture with USD operations. NVIDIA Omniverse fVDB NIMs create OpenUSD geometry

With fVDB's premiere at SIGGRAPH, OpenVDB's ten-year evolution continues, advancing the ways in which real-world digital twins could assist various businesses

NVIDIA Research, NVIDIA DRIVE, and NVIDIA Omniverse projects use it to portray huge, complex real-world settings with great fidelity