Técnicas para renderizado espectral en tiempo real

Pérez García, Pedro José
Monzón González, Néstor (dir.) ; Muñoz Orbañanos, Adolfo (dir.)

Universidad de Zaragoza, EINA, 2024
Departamento de Informática e Ingeniería de Sistemas, Área de Lenguajes y Sistemas Informáticos

Máster Universitario en Robótica, Gráficos y Visión por Computador

Resumen: Traditional RGB rendering presents problems during image synthesization upon encountering wavelength-dependent phenomena, such as participating media scattering or iridiscence. Spectral rendering takes into account the entire visual spectrum, and thus, is able to correctly evaluate these types of phenomena. However, RGB rendering is still the dominant choice due to how expensive it is to produce spectral materials and assets based on real-world data, and how heavy in memory those can be both in memory and disk storage, making the entire process of moving and storing that data a cumbersome task that introduces overhead and causes bottlenecks in real-time rendering engines, where computation time is critical. We present a physically-based, real-time spectral rendering pipeline that aims to mitigate these two factors by enabling the use of RGB assets in spectral contexts, introducing as little error as possible. For that end, we propose a real-time adaptation of a spectral upsampling technique, which allows for obtaining spectral responses for reflectances from the RGB coefficients in the original reflectance texture. To exemplify this proposed pipeline, we offer an implementation that aims to be performant, extensible and follows the main principles of physically based rendering. For this last goal, we allow for simulating different observer response curves and we implement physically based materials. We implemented our pipeline in OpenGL form scratch, excluding some libraries that alleviated the work on the most basic tasks. To prove the extendability of our approach, we combine it with previously existing techniques for real-time spectral rendering. We chose a technique to render underwater oceanic scenes in real time, allowing us to fully render those scenes in a spectral fashion. Lastly, we validate our results against a path traced simulation, showing great accuracy for wavelength-dependent scenarios and outperforming RGB rendering. Our method proves to be the most reliable across all the test scenarios, offering great frame rates for a little price in terms of processing power.

Tipo de Trabajo Académico: Trabajo Fin de Master