Dynamic Ray Allocation for Aliasing Mitigation in DDGI through Importance Sampling : MS-DDGI: Multisampling Dynamic Diffuse Global Illumination

Sammanfattning

Background. Modern video games rely on advanced rendering techniques to simulate realistic lighting. Probe-based global illumination systems relying on ray tracing like DDGI, a modern and dynamic approach introduced by Z. Majercik et al., can suffer from aliasing artifacts, particularly in scenes with small, bright emissive light sources. These artifacts arise from inconsistent and insufficient sampling of the environment around probes, leading to jarring visual patterns that degrade image quality and artistic flexibility. Objectives. This thesis aims to address the aliasing artifacts in DDGI by developing a novel technique called Multisampling Dynamic Diffuse Global Illumination (MS-DDGI). The primary goals are to investigate existing methods for artifact mitigation, design a dynamic importance sampling strategy for ray allocation, and evaluate the proposed solution's effectiveness in terms of visual quality, performance, and robustness. Methods. The research combines a literature review of Monte Carlo importance sampling and related rendering techniques with the implementation of MS-DDGI in the proprietary Snowdrop engine. The proposed system dynamically redistributes probe rays based on light importance, using probability density functions (PDFs) and cumulative distribution functions (CDFs) to prioritize high-impact directions. The evaluation includes qualitative visual comparisons, quantitative error analysis against reference renders, and performance profiling. Results. MS-DDGI significantly reduces aliasing artifacts, producing smoother and more consistent lighting in scenes with small emissive sources. Quantitative tests demonstrate lower mean squared error (MSE) compared to baseline DDGI, and domain experts confirm its visual improvements. However, the technique introduces a modest performance overhead in frame time and memory usage, with temporal delay as a notable trade-off. Conclusions. The study confirms that importance sampling can mitigate aliasing artifacts effectively for DDGI, enhancing visual fidelity and artistic usability. While the implementation incurs performance costs, the benefits justify its application in real-time rendering pipelines. Future work could optimize performance further and address temporal responsiveness.