Browse Teach Curate

Neural Constitutive Energy Models for 3D Garment Animation

How the mechanical behavior of cloth can be captured as a learned energy function (an Energy Unit network) trained directly on the observed motion of a single piece of fabric, encoding how stretching and bending change stored energy without any analytical model or differentiable simulator. How diverse 3D garments are then animated by optimizing their motion under this learned energy constraint, disentangling constitutive-behavior learning from garment-specific animation.

Dr. Theopolis Text

Learning 3D Garment Animation from Trajectories of A Piece of Cloth Yidi Shao1 Chen Change Loy1Bo

Garment animation for VR, games and film is usually driven by learning-based models trained per-garment on large motion datasets, which are expensive to collect and generalize poorly to unseen garmen…