A CPU-GPU hybrid computing framework for real-time clothing animation

Abstract

Real-time clothing animation has wide applications in many areas such as the design industry and e-commerce. Existing cloth simulation techniques based on the physical model can produce realistic animation effect. But they often incur a high computational load and hence are very time-consuming. GPU has become more widely used for high performance computing for its parallel processing capability. However, in the clothing animation problem, some tasks may not be suitable for GPU. In order to achieve the real-time realistic animation effect, this paper proposes a CPU-GPU hybrid computing framework. It is found that this framework can significantly improve bandwidth utilization. In this framework, the GPU is mainly in charge of high level parallel modules such as force calculation, collision detection and position update for each cloth vertex, while the CPU deals with the serial over-stretching process and other related computing. The implemented system achieves 10 times performance increase compared to the CPU mode, and can generate about 100 clothing animation key frames per second, thus meeting the real-time requirement for clothing animation.