This research aims to develop semantic-based compression techniques specifically designed for immersive 3D radiance field representations, focusing on improving efficiency while preserving the quality of experience (QoE). The objective is to integrate semantic analysis into the compression pipeline, leveraging scene semantics to identify and prioritize critical features in 3D data. This includes optimizing the representation and distribution of radiance field elements, such as splats or kernels, by adjusting their number, placement, and attributes based on the semantic importance of different regions in the scene. Additionally, the project will explore novel ways to couple semantic understanding and view synthesis, to improve rendering quality in dynamic and complex scenarios.This work will contribute to the development of scalable, generalizable compression solutions, advancing the capabilities of immersive technologies in applications like virtual reality, gaming, and telepresence.