PhysSkin++ learns real-time neural skinning without per-object training
Original title: PhysSkin++: Real-Time and Generalizable Physics-Based Animation via Self-Supervised Neural Skinning
A self-supervised autoencoder builds deformation subspaces from static geometry for meshes, clouds, and Gaussians.
PhysSkin++ learns continuous neural skinning fields from static geometries to form compact deformation subspaces. A Neural Skinning Fields Autoencoder predicts shape-dependent subspaces without training a network per object. Physics-informed self-supervision combines skinning-field normalization with conflict-aware gradient correction. The authors report cross-shape generalization, real-time reduced simulation, and controllable collision-aware animation.
Unconfirmed: Generalization and real-time claims are reported by the paper, not independently verified.
- Training does not require simulation trajectories.
- A complementarity objective learns elastic responses for prescribed rig animation.
- Complementarity constraints are not enforced at runtime.
- Animation can bind directly to meshes, point clouds, and 3D Gaussians.
From the original article
1 min read at huggingface.co
Generalizing real-time physics-based animation across diverse 3D shapes and spatial discretizations remains challenging. In this paper, we introduce PhysSkin++, a physics-informed framework that learns continuous neural skinning fields from static geometries to construct compact deformation subspaces.
By Yuanhang Lei, Baofei Liu, Xingxuan Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.