fVDB

Developed by NVIDIA, fVDB is an open-source deep learning framework for sparse, large-scale, high-performance spatial intelligence. It builds NVIDIA-accelerated AI operators on top of OpenVDB to enable reality-scale digital twins, neural radiance fields, 3D generative AI, and more. 

The fVDB PyTorch extension is available through the fVDB Early Access program. 

The fVDB Early Access program will begin accepting applicants in September 2024. 

Apply for Early Access Program
fVDB Github

fVDB 3D deep learning infrastructure enables large scale generative AI, city-scale Neural Radiance Fields (NeRFs), physics simulations, and generating meshes from 1 billion points. 

Generative Physical AI With Spatial Intelligence 

fVDB provides 3D deep learning infrastructure for massive datasets and high resolutions. It combines crucial AI operators into a single, coherent system built on the VDB format. fVDB is the infrastructure for generative physical AI with spatial intelligence. 

High Performance, High Resolution

fVDB AI operators are built on top of NanoVDB, which provides GPU-acceleration for OpenVDB. The framework supports operations like sparse convolution and ray tracing optimized for real time. fVDB enables faster training and real-time inference, minimizing memory footprint and maximizing data processing throughput.

Seamless Integration

If you’re already using the VDB format, fVDB can read and write existing VDB datasets out of the box. It interoperates with other libraries and tools, such as Warp for Pythonic spatial computing, and the Kaolin Library for 3D deep learning. Adopting fVDB into your existing AI workflow is seamless..


Explore Key Features 

Comprehensive Operator Set

fVDB provides differentiable operators, including convolution, pooling, attention, and meshing, all designed for high-performance 3D deep learning applications. These operators allow you to build complex neural networks for spatial intelligence, like large-scale point cloud reconstruction and 3D generative modeling.

Accelerated Ray Tracing

fVDB utilizes the Hierarchical Digital Differential Analyzer (HDDA) algorithm built into OpenVDB to deliver fast and accurate ray tracing. You can train neural radiance fields (NeRFs) at city-scale and generate ray-traced visualizations rapidly.

Optimized Sparse Convolution

fVDB’s sparse convolution operators can process massive 3D datasets. By optimizing memory access patterns and computational load, fVDB enables fast and high-accuracy spatial data processing, essential for tasks like volumetric data analysis and physics simulation. 

img-alt-text

Physics super-resolution.


Coming Soon: fVDB NIMs

Coming soon, fVDB functionality will be available as NVIDIA NIM inference microservices that enable developers to incorporate the fVDB core framework into USD workflows. fVDB NIMs generate OpenUSD-based geometry in NVIDIA Omniverse.

Learn How to Integrate Gen AI into your OpenUSD Workflow with NVIDIA USD NIM Microservices

More VDB Libraries

NanoVDB

Developed by NVIDIA, NanoVDB adds real-time rendering GPU support for OpenVDB. OpenVDB is the Academy Award-winning industry standard data-structure and toolset used for manipulating volumetric effects.

NeuralVDB

NeuralVDB is a large-scale volume representation with AI-enabled data compression technology. It offers significant efficiency improvements over OpenVDB, the industry-standard library for simulating and rendering sparse volumetric data such as water, fire, smoke, and clouds.


Resources

Join the fVDB Early Access Program

Apply Now