展开 为各种角色合成可信的运动一直是计算机图形学的目标。目前基于学习的运动合成方法依赖于大量的运动数据集，这些数据集往往很难获得，甚至是不可能获得的。另一方面，姿势数据更容易获得，因为静态姿势字符更容易创建，甚至可以使用计算机视觉的最新进展从图像中提取。我们利用这个替代数据源，并通过重定向引入了一种神经运动合成方法。我们的方法通过从另一个字符的现有运动捕获数据集转移运动，为只有姿态数据的字符生成合理的运动，该数据集可能具有截然不同的骨架。我们的方法有效地结合了源字符的运动特征和目标字符的姿态特征，并且在小的或有噪声的姿态数据集上表现稳健，从少数艺术家创建的姿态到直接从图像中估计的有噪声的姿态。返回搜狐，查看更多
has been the goal of computer graphics to launch a credible campaign for the synthesis of various roles . The current learning-based method of synthesizing sports relies on a large number of motion data sets, which are often difficult, if not impossible, to obtain. On the other hand, positional data is more readily available because static position characters are easier to create and even use computer visual advances to extract images. We use this alternative data source and have introduced a neuromotive synthesis method through re-direction. Our method is to create a different skeleton through a movement that captures data sets from an existing movement of another character and only posturing data, by creating a movement that is reasonable. Our method effectively combines the physical characteristics of the source character with the physical character of the object.
责任编辑：
注册有任何问题请添加 微信：MVIP619 拉你进入群
打开微信扫一扫
添加客服
进入交流群
发表评论