A Practical Implementation of Gradient Domain Path Tracing Rendering

Abstract

Gradient Domain Path Tracing is a modern technique under continuous improvement which has recently been published as a new rendering process. It studies an innovative approach in which a gradient image is constructed using Monte Carlo path tracing. The main contribution of this method consists of producing images with low variance and consequently reducing high frequency noise. After studying the current available results proposed by Kettunen et al., we aim to reproduce the same outcomes but through our own implementation of the algorithm. The first part of the work gives our interpretation and understanding of the available literature about this topic. The same theoretical approach will be summarily explained from our point of view, containing brief points about general physically based rendering. Afterwards, we will explain in depth our implementation of the Gradient Domain Path Tracer and the math behind it. In conclusion, we will discuss future work and improvements, possible drawbacks not considered in the paper and results comparison between our project and the outcomes of the original paper.