Deep Neural Networks for Geometric Shape Deformation

Geometric deep learning is a promising approach to bring the representational power of deep neural networks to 3D data. Explicit 3D representations such as point clouds or meshes can have varying and often a huge number of dimensions, what limits their use as an input to a neural network. Implicit representations such as signed distance functions (SDF) are on the contrary low-dimensional and fixed representations of the structure of a 3D shape that can be easily fed into a neural network. In this paper, we demonstrate how deep SDF neural networks can be used to precisely predict the deformation of a material after the application of a specific force. The model is trained using a set of custom finite element simulations in order to generalize to unseen forces.