Знешумлення з використанням машинного навчання для задачі трасування (PATH TRACING)

Abstract

This paper considers a method for denoising the image generated every frame after the path tracing step. The noisiness of the image after path tracing is inherent to the path tracing rendering (due to stochastic nature of selection of the vector that ray propagates along after hitting a rough surface). Thus denoising the image is essential for producing a visually appealing picture. Machine learning techniques for denoising are very promising due to their ability for many unapparent nuances. For instance which areas of the image require more processing, and which can make do with less. It could increase performance, which in turn allows for better picture rendered overall.