基于机器学习合成高分辨率布料褶皱  

Synthesis of High-resolution Cloth Folds Based on Machine Learning

作　　者：贾瑶 靳雁霞[1] 马博 陈治旭 芦烨 JIA Yao;JIN Yan-xia;MA Bo;CHEN Zhi-xu;LU Ye(School of Big Data,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学大数据学院,山西太原030051

出　　处：《计算机技术与发展》2021年第3期106-110,共5页Computer Technology and Development

基　　金：山西省自然科学基金(201801D121154,201801D121155)。

摘　　要：为了精化布料网格得到逼真的布料模拟效果,提出了一种基于机器学习的方法合成高分辨率布料褶皱。首先模拟真实的布料运动,获取布料运动的帧数据信息,将布料运动的帧数据信息转换为图像信息的形式进行存储。然后将图像信息作为输入,输入到卷积神经网络(CNN)中进行训练,通过将卷积神经网络与缩小网络相结合,最终得到高分辨率布料图像。最后将高分辨率布料图像转换为高分辨率布料网格,对布料进行模拟。实验结果表明,与初始的低分辨率布料网格对比,合成的高分辨率布料网格模拟出的布料有着大量且细微的褶皱,并且能够模拟出真实的布料效果,与真实场景中的布料模拟效果相似。该方法在不同的场景中都可以模拟出高质量的布料动画效果,而且减少了仿真速度,验证了该方法的有效性。In order to refine the cloth mesh and get a realistic cloth simulation effect,a method based on machine learning to synthesize high-resolution cloth folds is proposed.First,we simulate the real cloth movement,obtain the frame data information of the cloth movement,and convert the frame data of the cloth movement into an image for storage.Then the image information is input into the convolutional neural network(CNN)for training.By combining the CNN and the reduction network,a high-resolution cloth image is finally obtained.Finally,the high-resolution cloth image is converted into a high-resolution cloth grid to simulate the cloth.The experiment shows that compared with the original low-resolution cloth grid,the synthesized high-resolution cloth grid simulates a large number of fine folds,and can simulate the real cloth effect,similar to the cloth simulation effect in the real scene.The high-quality cloth animation effects of this method can be simulated in different scenes,and the simulation speed is reduced,which verifies its effectiveness.

关 键 词：机器学习 高分辨率 布料褶皱 卷积神经网络 布料动画 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]