Abstract
<title>Abstract</title> <p>This paper presents a topology-preserving anisotropic adaptive framework for yarn-level cloth simulation that maintains the woven structure. Conventional triangle-based adaptive cloth simulation can reduce computational cost through edge collapse, local remeshing, triangle simplification, and related operations. However, in yarn-level cloth, even the removal or merging of a single yarn can discontinuously alter the warp--weft crossing structure, contact relationships, and friction conditions, easily leading to popping or visual artifacts. To address this problem, this study introduces a direction-aware adaptive energy that independently evaluates deformation characteristics along the warp and weft directions without directly modifying the topology of yarns or nodes. The proposed method computes the importance of each region based on direction-dependent strain variation, curvature, sliding/contact activity, wrinkle tendency, and related factors. In regions with high energy, full constraint evaluation is maintained, whereas in regions with small changes, the intensity and frequency of constraint computations such as stretch, bending, shear, and collision are adaptively relaxed. In addition, hysteresis-based level switching, temporal filtering, minimum residence time, and force blending are applied to suppress frame-to-frame jitter and popping that may occur during adaptive transitions. Furthermore, while preserving the guide-yarn structure used in the simulation, a visual upsampling technique is applied at the rendering stage to interpolate and generate child yarns, providing a denser woven appearance without increasing computational cost. Experimental results on various weaving patterns and dynamic scenes demonstrate that the proposed method improves computational efficiency while preserving the woven structure and direction-dependent deformation characteristics of yarn-level cloth, producing stable visual results.</p>