Optimal-MIS

Practical implementation of Optimal MIS for Light Transport.

This repository provides a C++ header only library of Practical Optimal Multiple Importance Sampling, as presented in the TODO: Practical Optimal MIS technical report, based on the original Optimal MIS paper. We also implemented a path tracer and bidirectional path tracer in PBRT (https://github.com/SuperRonan/PBRT-Optimal-MIS) using this library.

The library provides two abstract classes, Estimator (in src/MIS/Estimator.h) and ImageEstimator (in src/MIS/ImageEstimator.h). Estimator is entended to be used by algorithms computing one pixel at once (like a path tracer), whereas ImageEstimator is intended to be used by algorithms for which the any pixel can be updated (like a bidirectional path tracer, for the contribution of the light tracer). The implementation of ImageEstimator is optimized over types like Image<Estimator>. The library uses the namespace MIS.

There are several implementation of Estimators for different weigthing functions

• Eric Veach's balance, power, cutoff, maximum heuristics in MIS::BalanceEstimator, MIS::PowerEstimator, MIS::CutOffEstimator,MIS::MaximumEstimator and MIS::ImageBalanceEstimator, MIS::ImagePowerEstimator, MIS::ImageCutOffEstimator,MIS::ImageMaximumEstimator respectively.
• A naive heuristic (1 / N) in MIS::NaiveEstimator and MIS::ImageNaiveEstimator.
• The direct estimator of the Optimal MIS weights in MIS::DirectEstimator and MIS::ImageDirectEstimator (slightly biased but efficient).
• TODO The progressive estimator of the Optimal MIS weights in MIS::ProgressiveEstimator and MIS::ImageProgressiveEstimator (unbiased, but far less efficient, both in runtime and variance reduction).

The files src/MIS/Estimators.h includes all the estimators, and src/MIS/ImageEstimators.h includes all the image estimators.

The estimators are templated by Spectrum and Float.

• Spectrum is intended to provide the functions described in Example/Spectrum.h, but can also be set to float or double.
• Float is intended to be float or double.

Using the library

To clone the library: git clone https://github.com/SuperRonan/Optimal-MIS.git --recurse-submodules. Do not forget the --recurse-submodules to also get Eigen.

Since this is a header only library, you need to add the src directory to the include directories of your project. For example with CMake, use: INCLUDE_DIRECTORIES(PathToOptimal-MIS/src). Optimal-MIS uses Eigen, another header only library for linear algebra. If Eigen is not already included, it is a submodule of this repository, under ext/Eigen. To include it with CMake, use: INCLUDE_DIRECTORIES(PathToOptimal-MIS/ext/Eigen). Make sure it is ext/Eigen that is added to your include directories, not ext/Eigen/Eigen. You can also do some CMake black magic I don't really know about (like FIND_PACKAGE(Eigen3 3.3 REQUIRED NO_MODULE) or something). Finaly, OptiMIS can benefit from OpenMP. It can easily be activated in CMake with:
FIND_PACKAGE(OpenMP)
SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
SET(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${OpenMP_EXE_LINKER_FLAGS}")