Abstract Tessellation is critical for real‐time rendering of Catmull‐Clark subdivision (CCS) surfaces, with control mesh singularities posing a major challenge. Prior works typically pre‐subdivide singular regions to a fixed depth offline to facilitate efficient runtime tessellation, yet this practice fails to meet view‐dependent error requirements for high‐fidelity rendering—especially at pixel‐level accuracy. Hence, we present EBOAT, a novel direct‐evaluation tessellation method that integrates memory‐efficient feature‐adaptive subdivision via a connectivity‐elided representation for singularities, along with a two‐phase error control strategy and improved subdivision depth estimation built upon it. With a uniform tessellation level of 64, EBOAT achieves a 3.8×–4.7× speedup over OpenSubdiv, without relying on fixed‐depth pre‐subdivision. Compared to state‐of‐the‐art breadth‐first subdivision, it yields a 4.2×–7.1× speedup while providing more accurate sampling. To further demonstrate its effectiveness, we implement a pixel‐accurate CCS surface rendering pipeline with EBOAT fully integrated, and provide an in‐depth evaluation of rendering quality. Given its advantages in both performance and accuracy, EBOAT offers a promising alternative for real‐time CCS surface rendering in industrial applications.
Zeng et al. (Fri,) studied this question.