Adaptive learning‐based projection method for smoke simulation. (May 2018)

• Record Type: Journal Article
• Authors: Xiao, Xiangyun; Yang, Cheng; Yang, Xubo
• Is Part Of: Computer animation and virtual worlds. Volume 29:Number 3/4(2018)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2018-05
• Subjects: deep neural network -- fluid simulation -- incremental learning; Computer animation -- Periodicals; Visualization -- Periodicals; 006.6
• DOI: 10.1002/cav.1837
• Languages: English
• ISSNs: 1546-4261

Abstract

Abstract: Traditional Eulerian‐based fluid simulations require much time and computational resources to solve the projection step, especially the large linear system produced by the Poisson equation. In this paper, we propose an adaptive machine‐learning‐based projection method combining deep neural network and incremental learning technique. We provide two modes: Fast Mode and Normal Mode to solve the most time‐consuming projection step and deal with various simulation scenes largely different from the training data set. We introduce patch‐based feature vectors to represent the whole velocity field and a loss function to keep the divergence‐free constraint. We have demonstrated the acceleration and extrapolation ability of our method by testing various smoke scenes far different from our training data set. Abstract : We provide two switchable modes. In the fast projection mode, we choose a deep neural network as the core pressure solver. In the normal projection mode, we employ traditional iterative methods as the Poisson solver to calculate pressure in the projection step and collect and save the training samples for DNN incremental training.