Shape representations for scalable 3D graphics pipelines

Recent research in the field of computer vision and graphics greatly widened the spectrum of available representations for 3D shapes, each offering distinct advantages depending on the application. Selecting the most suitable representation can have great impact on the efficiency and scalability of algorithms used throughout the computer graphics pipeline. This thesis aims to explore the influence of different 3D shape representations—ranging from traditional polygonal meshes to implicit, neural and hybrid representations—on the performance of modern graphics systems. Through the introduction of novel algorithms, we offer an overview of the key trade-offs between these representations via a detailed investigation of computational efficiency, memory footprint, and adaptability to different settings. Our case studies include geometry processing tasks, such as shape matching and editing, as well as rendering, texturing and simulation. Additionally, the thesis places particular emphasis on the critical role of data in recent research, offering four key data contributions that support the development and evaluation of scalable 3D shape representations. Empirical evaluations demonstrate the potential of this work to advance scalable 3D graphics pipelines, improving the rendering of large, dynamic scenes with high detail and interactivity.