Neural Radiance Caching

An implementation of Real-time Neural Radiance Caching for Path Tracing with OptiX 8 and tiny-cuda-nn.

Rendered in 2K resolution with an RTX 4080 Laptop GPU

How can I run this?

Tested on Windows with Visual Studio 2022.

Building for Linux, or Windows without VS, is possible, but you may have to manually set up the third-party dependencies.

Prerequisites

• OptiX SDK 7 or 8, and a supported NVIDIA GPU.
• CUDA Toolkit: Tested with CUDA 12.4, but any recent version compatible with your OptiX SDK should work.
• CMake 3.26 or higher.
• Visual Studio 2017 or higher.
• MDL SDK 1.8: Required for PBR materials. We recommend downloading the binary here.
• Additional dependencies (including tiny-cuda-nn) set up by the 3rdparty.cmd script.

Build & Run

We have included a CMakeLists.txt for the project and a 3rdparty.cmd script to install the additional dependencies.

You can build the project by following these steps:

• Launch the x64 Native Tools Command Prompt for VS 2017/2019/2022. (We will assume this as the shell environment for all commands below.)

• cd into the project root directory, and run 3rdparty.cmd.

• Set the environment variable MDL_SDK_PATH to point to your MDL library folder (containing an include folder.)

  • Alternatively, place your MDL library folder in the 3rdparty folder with the name MDL_SDK.

• To generate a solution file for the project, type

  mkdir build && cd build
  cmake -G "Visual Studio <version>" ..

  where <version> is 15 2017 for VS 2017, 16 2019 for VS 2019, and 17 2022 for VS 2022, resp.

To run the renderer,

• Set the working directory of the application to $(TargetDir) in Visual Studio (where the executable is located.)

• In the working directory, create a symbolic link to the data folder by typing

  mklink /D .\data <actual_data_folder>

  ⚠️ You might need admin privilege for this on Windows :\ (at least sudo is coming soon...)

• Try out the Cornell Box (with friends! 🐉🐇) scene with the command line arguments:

  -s data/system_mdl_cornell.txt -d data/scene_mdl_cornell_friends.txt

Demo

demo_cornell.mp4

demo_watercolor.mp4

Acknowledgement

Path tracer and assets dervied from the OptiX MDL renderer sample.