Deformable Objects Modeling Based On Neural Network

Physically based simulation is a popular research topic in computer graph-ics,which includes the simulation of rigid objects,deformable objects,fluid and so on And the physical simulation of deformable objects is a key to animation,movie making,medical treatment,and manufacturing.This technique has a large number of applications among the industries.The material model of deformable objects describe the relationship between deformation and force,which could be highly nonlinear.To satisfy the requirements of real material inference and ani-mation design scenarios,we learn a neural network to compensate a basic elastic model.In this work we propose a framework for learning models of deformable materials from sparse example surface trajectories,which predict deformable ob-jects'deformation.The challenge is that the surface trajectories is sparse as it is typically available only part of the surface.With this problem,we use sparse reduced space-time optimization identifies gentle correction forces with which we iteratively refine a radial basis function.Finally,the radial basis function con-verts to a neural network,which is a compact form to use in application step.Also,we demonstrate our method with a set of synthetic examples,as well as with data captured from real world homogeneous elastic objects.