Uppsats

Comparing Radiance Cascades to Voxel Cone Tracing for Interactive Global Illumination in Dynamic Scenes

Master-uppsats

Linköpings universitet/Informationskodning

Publicerad: 2025

Språk: Engelska

Sammanfattning

The importance of global illumination for photorealistic renderings in interactive applications drives the interest in algorithms that can efficiently calculate many-to-many light interactions in real time. This thesis explores Radiance Cascades, a new technique for global illumination, that aims to efficiently calculate and store light information in a hierarchical manner. This thesis implements and evaluates the performance of Radiance Cascades in an interactive application, and compares the visual results to those of the similar state-of-the-art technique Voxel Cone Tracing. This evaluation and comparison use different sampling resolutions, measured as the number of screen pixels between sample points, to look at a trade-off between performance and visual quality. The performance analysis shows an average render time of 5 ms, 20 ms and 30 ms for scenes of increasing size when sampling every fourth screen pixel. When sampling every other pixel, the average render times are 19 ms, 67 ms and 101 ms. Sampling every pixel yields too high render times to be feasible for real time applications. Although the render times are relatively stable, they each have a few significant large peaks. The visual comparison shows that the Radiance Cascades implementation suffers from light leaking at thin objects just like Voxel Cone Tracing, although to a lesser degree. Unlike Voxel Cone Tracing, the implementation also displays light leakage at depth discontinuities, visible individual rays around small light sources and flickering during object or camera movement. Additionally, the results indicate a trade-off between visual quality and render time as a function of the sampling resolution.

Information

Författare
Kung, Johannes
Lärosäte / institution
Linköpings universitet/Informationskodning
Publiceringsdatum
2025
Uppsatstyp
Master-uppsats
Språk
Engelska

Utforska vidare

Liknande uppsatser

Uppsatser med liknande ämnen och nyckelord.