An implementation of NeRF acceleration using RTX cores to compute ray-grid intersections
You will need:
- a Nvidia GPU with compute capability>7.0 and driver version >=530.41.
- Optix 7.7
- CUDA
- CMake > 3.18
In order to setup the repository, first download Optix. Then clone the repo. You'l first need to setup submodules.
git submodule init
git submodule update --init --recursive
If your Linux doesn't already have them installed install the following:
sudo apt-get install build-essential
We then need to build tiny-cuda-nn (this takes time)
cd lib/tiny-cuda-nn
mkdir build && cd build
cmake ..
make -j
Now that the tiny-cuda-nn static library is built we can compile the rtx_nerf project.
Navigate to the project root, and run the following:
mkdir build && cd build
cmake ../ -DOPTIX_HOME=[/path/to/optix/]
make
In order to build in debug mode
cmake ../ -DCMAKE_BUILD_TYPE=Debug -DOPTIX_HOME=[/path/to/optix/]
make
First download the data from the NeRF website. Then create a data folder and extract the data.
From the project root, run the executable
./build/bin/optx_nerf