Implementation Tutorial for Neural Radiosity [Hadadan et al. 2021]

See the python notebook neural_radiosity.ipynb.

(Almost) Everything about installing dependencies

Installation

Test environment

• Ubuntu 20.4
• NVIDIA driver version 515.86.01
• CUDA driver version 11.7
• CUDA runtime version 11.7
• GCC >= 8
• CMake >= 3.21

git clone https://github.com/krafton-ai/neural-radiosity-tutorial-mitsuba3.git
python3 -m pip install --upgrade pip
pip install -r requirements.txt

To install Tiny-CUDA-NN, check the official documentation. In short,

# but first, you may need to update CUDA toolkit and CMake as the following sections
apt-get install build-essential git python3-dev python3-pip libopenexr-dev \ 
    libxi-dev libglfw3-dev libglew-dev libomp-dev libxinerama-dev libxcursor-dev \ 
    ffmpeg
pip install "git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch"

When you encounter 'no LLVM.so' error while using 'llvm_ad_rgb' variant of Mitsuba3

# if you encounter something like 'no LLVM.so' error,
apt-get update
apt-get -y install llvm
find / -iname libLLVM.so  # verify the installation
pip uninstall mitsuba drjit
pip install -r requirements.txt

Installation on WSL

1. Go to Microsoft Store and install Windows Subsystem for Linux (WSL). See here for more detailed information.
2. Go to Microsoft Store and install VSCode.
3. Open a VSCode window, and install the WSL extension from VSCode. See here if you are not aware of VSCode Extensions.
4. In a VSCode window, press ctrl+shift+p and write WSL to find WSL: Connect to WSL option. You can now open a VSCode window on a virtualized Ubuntu environment by pressing the option. See here if you are not aware of the ctrl+shift+p command in VSCode.
5. The remaining steps are the same as in the Installation section above.

Update CUDA Toolkit

• For instance, Tiny-CUDA-NN requires a recent verion of CUDA toolkit which is 11.5 or higher.
• To install, CUDA toolkit 11.7, follow the link
• You can also find other versions of toolkits in the NVIDIA archive
• Anyway the above 11.7 link will introduce the following installation instructions:

wget https://developer.download.nvidia.com/compute/cuda/11.7.0/local_installers/cuda_11.7.0_515.43.04_linux.run
sh cuda_11.7.0_515.43.04_linux.run

# 1. accept the license stuff
# 2. uncheck `Driver`, `CUDA Samples xx.x`, `CUDA Demo Suite xx.x`, `CUDA Documentation  xx.x`.
#    we only need 'CUDA Toolkit xx.x`

• After the installation is finished, open ~/.bashrc and add some paths as follow.

...
export PATH=/usr/local/cuda-11.7/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.7/lib64

Update CMake

• Get the CMake binary for Linux such as cmake-3.26.0-rc6-linux-x86_64.sh.
• Install the binary by hitting sh cmake-3.26.0-rc6-linux-x86_64.sh.
• Add some environment variables

...
export PATH=~/cmake-3.26.0-rc6-linux-x86_64/bin:$PATH
export CMAKE_PREFIX_PATH=~/cmake-3.26.0-rc6-linux-x86_64:$CMAKE_PREFIX_PATH

LICENSE

The code is released under the Apache-2.0 License. See LICENSE for full terms.