Uppsats
Temporal Anti-Aliasing for Real-Time Froxel-Based Volumetric Rendering - Advances in volumetric temporal anti-aliasing for real-time ren dering of participating media
Master-uppsats
Göteborgs universitet/Institutionen för data- och informationsteknik
Publicerad: 2026-06-30
Språk: Engelska
Sammanfattning
Froxel-based rendering is widely used in modern graphics pipelines to render participating media such as fog and atmospheric lighting. Due to the low resolution ofthe froxel grid, required for real-time rendering feasibility, these techniques sufferfrom temporal instability, including flickering, shimmering and ghosting artifacts.Existing temporal accumulation methods that use an exponential moving averageprovide stable images in static scenes but perform poorly under dynamic volumetriclighting conditions.This thesis investigates how temporal anti-aliasing techniques originally developed forsurface-based rendering can be extended into froxel space for volumetric renderingof participating media. Several history validation, blending, jittering, filtering,and spatial reconstruction methods were implemented and evaluated in Godot andFrostbite. The methods were compared quantitatively against a supersampledvolumetric ground truth and qualitatively in dynamic lighting conditions.The results show that variance clipping applied directly to volumetric froxel data,together with heuristic anti-flicker techniques, greatly reduces ghosting artifactswhile introducing only a small amount of flickering. By considering the six nearestfroxels when computing current-frame local variance estimates, the best trade-offbetween temporal stability, spatial accuracy and performance cost is achieved. Depthonly low-discrepancy jittering and weighting samples based on their distance to thefroxel center further helps reduce flickering artifacts. The proposed volumetricTAA algorithm was successfully implemented in Frostbite, showing clear visualimprovements and acceptable runtime performance on modern console hardware.
Information
- Författare
- Lerviks, Birk
- Lärosäte / institution
- Göteborgs universitet/Institutionen för data- och informationsteknik
- Publiceringsdatum
- 2026-06-30
- Uppsatstyp
- Master-uppsats
- Språk
- Engelska