Month: July 2019

Daniel Holden, Bang Chi Duong, Sayantan Datta, Derek Nowrouzezahrai

Data-driven methods for physical simulation are an attractive option for interactive applications due to their ability to trade precomputation and memory footprint in exchange for improved runtime performance. Yet, existing data-driven methods fall short of the extreme memory and performance constraints imposed by modern interactive applications like AAA games and virtual reality. Here, performance budgets for physics simulation range from tens to hundreds of micro-seconds per frame, per object. We present a data-driven physical simulation method that meets these constraints. Our method combines subspace simulation techniques with machine learning which, when coupled, enables a very efficient subspace-only physics simulation that supports interactions with external objects – a longstanding challenge for existing sub-space techniques. We also present an interpretation of our method as a special case of subspace Verlet integration, where we apply machine learning to efficiently approximate the physical forces of the system directly in the subspace. We propose several practical solutions required to make effective use of such a model, including a novel training methodology required for prediction stability, and a GPU-friendly subspace decompression algorithm to accelerate rendering.

Subspace Neural Physics: Fast Data-Driven Interactive Simulation

Peter Kadlecek, Ladislav Kavan

The human face is an anatomical system exhibiting heterogenous and anisotropic mechanical behavior. This leads to complex deformations even in a neutral facial expression due to external forces such as gravity. We start by building a volumetric model from magnetic resonance images of a neutral facial expression. To obtain data on facial deformations we capture and register 3D scans of the face with different gravity directions and with various facial expressions. Our main contribution consists in solving an inverse physics problem where we learn mechanical properties of the face from our training data (3D scans). Specifically, we learn heterogenous stiffness and prestrain (which introduces anisotropy). The generalization capability of our resulting physics-based model is tested on 3D scans. We demonstrate that our model generates predictions of facial deformations more accurately than recent related physics-based techniques.

Building Accurate Physics-Based Face Models From Data