Steps

• Setup WSL2 with CUDA support

• Install CUDA Toolkits via conda

conda create -n instant-ngp
conda activate instant-ngp
conda install -c nvidia cuda-tools cuda-toolkit
conda install -c conda-forge cmake

export PATH="$CONDA_PREFIX/bin:$PATH"
export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"

• Install other dependencies via apt

sudo apt-get install build-essential git python3-dev python3-pip libopenexr-dev libxi-dev \
                     libglfw3-dev libglew-dev libomp-dev libxinerama-dev libxcursor-dev

• Clone the repo from Github

git clone --recursive https://github.com/nvlabs/instant-ngp
cd instant-ngp

• Modify the source code to fix the SegmentationError

GLTexture::CUDAMapping::CUDAMapping(GLuint texture_id, const Vector2i& size) : m_size{size} {
	// static bool s_is_cuda_interop_supported = true;
	static bool s_is_cuda_interop_supported = false;
	if (s_is_cuda_interop_supported) {
		cudaError_t err = cudaGraphicsGLRegisterImage(&m_graphics_resource, texture_id, GL_TEXTURE_2D, cudaGraphicsRegisterFlagsSurfaceLoadStore);
		if (err != cudaSuccess) {
			s_is_cuda_interop_supported = false;
			cudaGetLastError(); // Reset error
		}
	}

	if (!s_is_cuda_interop_supported) {
		// falling back to a regular cuda surface + CPU copy of data
		m_cuda_surface = std::make_unique<CudaSurface2D>();
		m_cuda_surface->resize(size);
		m_data_cpu.resize(m_size.prod() * 4);
		return;
	}

	CUDA_CHECK_THROW(cudaGraphicsMapResources(1, &m_graphics_resource));
	CUDA_CHECK_THROW(cudaGraphicsSubResourceGetMappedArray(&m_mapped_array, m_graphics_resource, 0, 0));

	struct cudaResourceDesc resource_desc;
	memset(&resource_desc, 0, sizeof(resource_desc));
	resource_desc.resType = cudaResourceTypeArray;
	resource_desc.res.array.array = m_mapped_array;

	CUDA_CHECK_THROW(cudaCreateSurfaceObject(&m_surface, &resource_desc));
}

• Build

cmake . -B build
cmake --build build --config RelWithDebInfo -j 16

• Run

./build/testbed --scene data/nerf/fox