高级检索
周宇明, 苏科华. 使用GPU渲染的离散最优传输算法[J]. 计算机辅助设计与图形学学报, 2019, 31(5): 726-735. DOI: 10.3724/SP.J.1089.2019.17375
引用本文: 周宇明, 苏科华. 使用GPU渲染的离散最优传输算法[J]. 计算机辅助设计与图形学学报, 2019, 31(5): 726-735. DOI: 10.3724/SP.J.1089.2019.17375
Zhou Yuming, Su Kehua. A GPU Rendering Algorithm for Discrete Optimal Mass Transportation[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(5): 726-735. DOI: 10.3724/SP.J.1089.2019.17375
Citation: Zhou Yuming, Su Kehua. A GPU Rendering Algorithm for Discrete Optimal Mass Transportation[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(5): 726-735. DOI: 10.3724/SP.J.1089.2019.17375

使用GPU渲染的离散最优传输算法

A GPU Rendering Algorithm for Discrete Optimal Mass Transportation

  • 摘要: 针对离散最优传输算法复杂实现难度大的问题,将最优传输转换成多个三维平面的渲染问题,提出一种利用GPU渲染管线以绘制四边形的方式求解的简单算法.首先根据最优传输的原像计算得到一系列三维空间中的平面;然后使用正交相机对这些平面进行渲染得到其垂直投影,并根据投影中每个胞腔的面积可以得到当前测度;接着使用梯度下降法调整平面的位置,使得当前测度等于目标测度,得到最优传输的结果;最后基于该算法构建了拓扑圆盘网格的保面积参数化算法.使用Maxplanck, Alexraw, Lion, Totoro和Buddaha模型进行实验,与使用数值法进行比较,该算法的迭代速度提升了8倍;与其他类似的算法进行对比,使用面积之比取对数作为评判指标,结果表明该算法的保面积效果更好.

     

    Abstract: To solve the problem of difficulty in implementing the discrete optimal transportation algorithm,this paper presents a simple algorithm that solve the problem by rendering quadrilateral with the help of GPU rendering pipeline. Firstly, a series of 3D plane was built from the preimage of optimal mass transportation map. In order to obtain the vertical projection, these planes were rendered using orthographic camera.Current measure can be computed from it. Then, gradient descent was used to adjust the planes’ position so as to let current measure be equal to target measure, obtaining the results of optimal mass transportation.Finally, an equiareal parameterization for topological disk was constructed based on the algorithm The algorithm was evaluated on the models of Maxplanck, Alexraw, Lion, Totoro and Buddaha. Compared with the numerical method, the iteration speed of this algorithm is improved by 8 times and compared with other similar methods, taking the logarithm of ratio of area as evaluation criteria, our algorithm is more area-preserving.

     

/

返回文章
返回