To address interpenetration problems in neural garment prediction, we propose a novel collision handling neural network layer called Repulsive Force Unit (ReFU). Based on the signed distance function (SDF) of the underlying body and the current garment vertex positions, ReFU predicts the per-vertex offsets that push any interpenetrating vertex to a collision-free configuration while preserving the fine geometric details. Our experiments show that ReFU significantly reduces the number of collisions between the body and the garment and better preserves geometric details compared to prior methods based on collision loss or post-processing optimization.
We address the problem of accelerating thin-shell deformable object simulations by dimension reduction. We present a new algorithm to embed a high-dimensional configuration space of deformable objects in a low-dimensional feature space, where the configurations of objects and feature points have approximate one-to-one mapping.
Author
Qingyang Tan (qytan@cs.umd.edu)
Abstract
In this project, we want to build a cloth simulator in a nonlinear subspace to reduce the search space. We leverage the graph-based convolutional neural network (GCNN) with a physical-inspired loss term to build an embedding network for cloth models. We then use mass-spring systems or finite-element method (FEM) to model the cloth energy and build a simulator based on the nonlinear embedded space. We show that using the proposed method can better capture cloth deformation and generalize on new control information well.