A garment acquisition pipeline for game asset retargetting

点击翻译按钮获取中文摘要

Cloth animation is ubiquitous in video games and a quintessential part of game playing experience since the dawn of computer gaming. Realistic cloth animation can be achieved using sophisticated cloth simulation engines, but in a real-time game the computing budget for cloth simulation simply does not allow for a detailed cloth simulation. Therefore, practical solutions generally revolve around augmenting simplified simulation techniques: a simple simulation at the base augmented by keyframing or other animation techniques for increased animator control. This approach has a number of shortcomings from lack of animator control, to simulation stability as well as performance. In this work we propose a fundamentally different approach where we use off-line cloth acquisition to drive the cloth synthesis process in a homologue way to how motion capture can drive the character motion synthesis process. We use performance captures of cloth as assets to be retargeted in real-time based on a never seen before character and motion. We achieve this by using the recent advances in artificial intelligence and neural networks. We define the animation of a specific garment as a four-dimensional regression function of body shape, body dynamics and garment properties and we frame the problem as an optimization function where the input is body rest shape, body motion, garment rest-shape and the output. We are collaborating closely on this project with UBISOFT, a leading company in computer games that has a large R&D studio in Montreal. Using this technique will achieve a significant paradigm shift in computer animation that has the potential of revolutionizing the computer gaming field thus benefiting the company, the computer animation community as well as the Canadian gaming industry as a whole.