Progressive Consistent Computation of Global Illumination

Dean's award for outstanding master thesis

Progressiveness in a global illumination algorithm is a desirable, yet seldom considered property. We have decided to explore this class of algorithms with additional focus on estimate consistency and suitability for GPU implementation. Of these, several were chosen for implementation, including path tracing and various filtering techniques on one hand and several progressive photon tracing algorithms on the other. In total, seven algorithms were implemented. The process of parallelizing these for a Single Program Multiple Data (SPMD) parallel computer is described, the resulting implementations were then tested on several scenes with varying complexity and illumination features. Results were compared, with main focus on progression of estimation error in time.