Real-time XR visualization of 3D Gaussian Splatting models

The demand for high-quality, immersive experiences in eXtended Reality (XR) environments has increased significantly, driven by applications in entertainment, education, healthcare, robotics and industrial design. XR technologies rely heavily on real-time rendering of 3D scenes to create interactive and responsive experiences. However, achieving both visual fidelity and computational efficiency remains a major challenge, particularly on mobile and wearable devices with limited resources. Emerging approaches, such as 3D Gaussian Splatting, are transforming how 3D scenes are visualized in real time. These innovations offer the potential to enhance user experiences and expand the practical applications of XR technologies across various domains. This thesis explores the application of 3D Gaussian Splatting in real-time XR visualization, proposing an adaptive Level of Detail system to optimize rendering performance. By dynamically adjusting scene complexity based on user position and the importance of objects within the scene, this approach seeks to enhance both visual quality and computational efficiency. Semantic segmentation is leveraged to prioritize and allocate resources effectively, ensuring that detailed rendering is focused on key areas while less important regions are simplified. The integration of these techniques allows for high-quality visualizations that adapt to real-time requirements, enabling for XR applications that balance computational constraints with the need for high visual fidelity.