Reproduce “PhySkin: Physics-based Bone-driven Neural Garment Simulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Astitva Srivastava, Hsiao-yu Chen, Ryan Goldade, Philipp Herholz, Zhongshi Jiang, Gene Wei-Chin Lin, Lingchen Yang, Nikolaos Sarafianos and 2 more
PhySkin introduces a self-supervised neural framework for physics-based quasi-static garment simulation, achieving microsecond-level inference on a single-threaded CPU while producing physically plausible drapes that generalize across diverse body shapes, poses, and garment topologies. This enables real-time, high-quality garment animation on computationally constrained devices.
Key insights
- Decoupling body shape and pose processing via a modulation strategy allows for an extremely efficient neural network architecture that achieves microsecond-level inference on a single-threaded CPU.
- Physics-based self-supervised training effectively achieves physically plausible garment drapes, demonstrating quality comparable to supervised methods without requiring expensive pre-computed simulation data.
- Using a small set of affine 'bone handles' to drive garment deformation, combined with neural correction, provides a robust and generalizable solution across diverse garment topologies and resolutions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.