Self-supervised Garment Dynamics with Persistent Wrinkles
For computer graphics and animation, this work enables more realistic garment simulation without training data, addressing a key limitation in self-supervised neural simulators.
This paper introduces the first self-supervised neural garment simulator that models persistent wrinkles caused by plasticity, using a novel physics-inspired loss function and curriculum learning. It outperforms existing methods on various garments, body shapes, and motions, generating natural persistent wrinkles.
The self-supervised neural garment simulator has become popular due to its high efficiency, good visual realism, and no reliance on training data. However, existing methods greatly simplify the mechanical properties of fabrics, ignoring persistent wrinkles caused by plasticity. Although this simplification allows for modeling of purely elastic material and simple training via energy minimization, the lack of believable wrinkles adversely affects the visual realism. Therefore, we introduce the first self-supervised neural garment simulator that explicitly models persistent wrinkles. This is achieved by a novel physics-inspired loss function, turning learning into a moving energy minimization problem to mimic plasticity. However, this requires learning to use a changing loss function, which causes difficulties in training i.e. the loss function changes during training. To this end, we propose a new physics-inspired curriculum learning scheme where the target material for learning gradually changes from pure elasticity to elasto-plasticity, allowing the loss function and the learnable parameters to jointly converge. Through a comprehensive evaluation, we show that for the first time, self-supervised learning models can generate natural persistent wrinkles, outperforming existing methods on a variety of garments, body shapes, and body motions, according to a range of metrics.