Real-time path-tracing denoising

Details

Supervisors

De Vleeschouwer, Christophe ; Legrand, Antoine

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en sciences informatiques, à finalité spécialisée

Abstract

Path-Tracing techniques are at the heart of many tools and frameworks used to generate images, thereby enabling unified physically-based rendering pipelines. But given the current computing power and real-time requirements, the result is often noisy, due to the low number of samples evaluated. A new pass, which we will define as denoising, must be developed to allow the production of a high-quality image. This master thesis will evaluate and compare different techniques based on deep learning, filtering and regression. With predefined metrics, we consider both their effectiveness and computational efficiency. Additionally, this work will explore the integration of these denoising techniques in a more general rendering engine, identifying the dependencies required for their effective performance.