Neural Radiance Fields (NeRF) in Universal Scene Description (OpenUSD)

Demo Video

A comprehensive interface to solve data collection for Neural Radiance Fields and Novel View Synthesis for the Pixar OpenUSD framework.

Requirements

• CMake
• A local build of USD
• Qt6
• Python ~3.9

Custom Hydra Render Engine and Data Collection GUI

Build Instructions

Locate the helper script ./ez_build.sh found at the root of this repository.

Replace cmake_install_prefix with the repository path.

Replace the -DUSD_ROOT CMake flag with the path to your USD build.

Run ./ez_build.sh, followed by ./ez_run.sh (or alternatively just locate the executable within your build).

NeRF Python program with PyTorch and NumPy:

Build Instructions

Activate your choice of Python virtual environment (or not) and install python dependencies.

pip install -r src/nerf/requirements.txt

Run the main Python module.

python src/nerf/main.py

Note: if you choose to run from another directory, you must modify any path elements in all Python modules located in src/nerf/.