Installation

An Nvidia graphics card is necessary to run this software. In addition to that the following software packages will also be necessary.

Install Nvidia driver, Cuda and Optix 7. Please follow this blog post blog post to install these packages.
Install OWL (A Node Graph "Wrapper" Library for OptiX 7). Please follow the Building OWL instructions provided on OWL github repository OWL github repository
Eigen3 3.3

sudo apt-get update
sudo apt-get install libeigen3-dev

sudo apt-get update
sudo apt-get install nlohmann-json3-dev

OpenCV 4.5.1 with GPU support. Build and install this using the instructions provided in this OpenCV blog post

Build

git clone --recurse-submodules https://github.com/DRealArun/owl-sphere-test

cd owl-sphere-test
mkdir data
mkdir models

Copy the blender simulation dataset to the data directory (Ensure that the depth maps, images and .json file are in the data directory)
Download the monocular depth estimation models (links below) into the models directory that we created,
- MiDaS v2.1 Small : https://github.com/isl-org/MiDaS/releases/download/v2_1/model-small.onnx
- MiDaS v2.1 Large : https://github.com/isl-org/MiDaS/releases/download/v2_1/model-f6b98070.onnx
Build by running the following commands

mkdir build
cd build
cmake ..
make

Update the OpenCV_DIR in the CMakeLists.txt to point to your OpenCV installation (Folder that contains the OpenCVConfig.cmake and other .cmake files).
Enable openexr support at runtime by running the following command in the terminal,

export OPENCV_IO_ENABLE_OPENEXR=1 # This sets the enviroment variable used by opencv

To run the application on simulation blender data using groundtruth depth maps, run the following command.

./sphere-test

To run the application on simulation blender data using MiDaS v2.1 Small model, run the following command.

./sphere-test --infer-depth-s

To run the application on simulation blender data using MiDaS v2.1 Small model, run the following command.

./sphere-test --infer-depth-l