Abstract : Cloth simulation is widely used in real-time interactive applications; however, yarn-level simulation based on actual fiber structures is difficult to apply in real-time environments, such as games and virtual reality, due to its high computational cost. Existing approaches either rely on texture- or shader-based visual detail enhancement, or require physically based fiber-level simulations, both of which have inherent limitations. In this paper, we propose a lightweight framework that reconstructs yarn-level appearance and motion in real time using only the results of Position-Based Dynamics (PBD) cloth simulation, without explicit yarn modeling or fiber dynamics simulation. The proposed method decomposes cloth surface deformation into local tangential and normal components, and applies directional deformation filtering with anisotropic weights to effectively reproduce the directional deformation behavior of knitted structures. In addition, velocity-driven UV slip and a micro-deformation model are combined to stably reproduce subtle oscillation and twisting effects characteristic of fiber bundles. The method is integrated with tube-based geometric generation using Parallel Transport Frames, allowing the yarn resolution to be increased independently of the physical simulation, and enabling real-time rendering of high-resolution knitted structures. Experimental results demonstrate that the proposed method provides visually enhanced fiber-level appearance compared to conventional cloth surface rendering, while maintaining real-time frame rates. The proposed approach presents an efficient framework for yarn-level appearance synthesis that is suitable for real-time applications without requiring additional fiber dynamics computation.

[paper]