Português English

Tese de Doutorado de Jonas Deyson Brito dos Santos

Detalhes do Evento

Aluno: Jonas Deyson Brito dos Santos
Orientador: Prof. Dr. Manuel Menezes de Oliveira Neto

Título: A Framework for Developing and Benchmarking Sampling and Denoising Algorithms
Linha de Pesquisa: Computação Gráfica e Visualização de Dados

Data: 08/11/2019
Horário: 09h30
Local: Sala 215 (sala de videoconferência) do Prédio 43412 do Instituto de Informática da UFRGS

Banca Examinadora:
Prof. Dr. Eduardo Simões Lopes Gastal (UFRGS)
Prof. Dr. Gladimir Valério Guimarães Baranoski (University of Waterloo – por videoconferência)
Prof. Dr. Waldemar Celes Filho (PUC-Rio – por videoconferência)

Presidente da Banca: Prof. Dr. Manuel Menezes de Oliveira Neto

Abstract: In the context of Monte Carlo rendering, although many sampling and denoising techniques have been proposed in the last few years, the case for which one should be used for a specific scene is still to be made. Moreover, developing a new technique has required selecting a particular rendering system, which makes the technique tightly coupled to the chosen renderer and limits the amount of scenes it can be tested on. In this work, we propose a renderer-agnostic framework for developing and benchmarking sampling and denoising techniques for Monte Carlo rendering. It decouples techniques from rendering systems by hiding the renderer details behind a general API. This improves productivity and allows for direct comparisons among techniques using scenes from different rendering systems. The proposed framework contains two main parts: a software development kit that helps users to develop and and test their techniques locally, and an online system that allows users to submit their techniques and have them automatically benchmarked on our servers. We demonstrate its effectiveness by using our API to instrument four rendering systems and a variety of Monte Carlo denoising techniques — including recent learning-based ones — and performing a benchmark across different rendering systems.

Keywords: Monte carlo rendering, adaptive sampling and reconstruction, denoising, benchmark.