This is a simple implementation of NVidia AI denoiser. You can find a pre-built windows distribution either on my website here or in the releases tab of this repro. To build you will need to install the CUDA took availible from here and the OptiX 7/8 SDK availible here.
You will require an Nvidia driver of at least 465.84 or higher and an Nvidia GPU of Maxwell architecture or newer to use the OptiX denoiser.
You need to set the OPTIX80_PATH
environment variable, which is your OptiX SDK Path(eg /home/rabbit/NVIDIA-OptiX-SDK-8.0.0-linux64-x86_64).
sudo apt install -y openimageio-tools
mkdir build && cd build
cmake ..
make
Command line parameters
You need to at least have an input and output for the app to run. If you also have them, you can add an albedo AOV or albedo and normal AOVs to improve the denoising. All images should be the same resolutions, not meeting this requirement will lead to unexpected results (likely a crash).
For best results provide as many of the AOVs as possible to the denoiser. Generally the more information the denoiser has to work with the better. The denoiser also prefers images rendered with a box filter or by using FIS.
Here is a quick example scene that uses the images that can be found in the image folder of this repository.
As it has been widely requested here is a very simple batch script for denoising sequences until I have time to implement something proper into the application itself. It will do the most simple denoising without any feature AOVs. Save the following code into a file named Sequence.bat and place it into the directory where your images are saved. Running this script will denoise all files image files that match the chosen file extension in the folder. There are three parameters that you will need to edit in the script,
SET FILE_EXTENSION=jpg
SET PATH_TO_DENOISER=D:\Projects\NvidiaAIDenoiser\Denoiser_v2.0
SET OUTPUT_PREFIX=denoised_
for /r %%v in (*.%FILE_EXTENSION%) do %PATH_TO_DENOISER%\Denoiser.exe -i "%%~nv.%FILE_EXTENSION%" -o "%OUTPUT_PREFIX%%%~nv.%FILE_EXTENSION%"
cmd /k
This project is shared under the MIT License.