Real-Time Cloth Rendering with Fiber-Level Detail.

Area of Science:

• Computer Graphics
• Geometric Modeling
• Real-time Rendering

Background:

• Fiber-level cloth modeling offers high realism but incurs substantial memory and computation costs.
• Existing methods are often too slow for real-time applications, even for offline rendering.

Purpose of the Study:

• To develop a real-time fiber-level cloth rendering method suitable for modern Graphics Processing Units (GPUs).
• To significantly reduce the computational and memory overhead associated with detailed cloth rendering.

Main Methods:

• Procedurally generating fiber-level details on-the-fly using yarn-level control points to minimize GPU data transfer.
• Reducing rasterization by collectively representing fibers within yarn structures and employing level-of-detail (LOD) strategies.
• Implementing efficient self-shadow computation using low-resolution shadow maps and approximating ambient occlusion for fiber-level self-occlusion.

Main Results:

• Demonstrated real-time rendering of knitwear with over 100 million fiber curves.
• Achieved effective self-shadowing and ambient occlusion effects at real-time frame rates.
• Showcased the method's ability to handle physical-based shading and temporal coherency in cloth animations.

Conclusions:

• The proposed method enables real-time, high-fidelity rendering of fiber-level cloth.
• It effectively balances visual detail with computational efficiency for interactive applications.
• The approach is scalable and applicable to complex textile simulations and visualizations.