← ArchivePaper2022
Neural Cloth Simulation
First unsupervised deep learning framework for garment dynamics using physics-inspired losses without ground-truth simulation data.
Abstract
We present a general framework for the garment animation problem through unsupervised deep learning inspired in physically based simulation. Existing trends in the literature already explore this possibility. Nonetheless, these approaches do not handle cloth dynamics. Here, we propose the first methodology able to learn realistic cloth dynamics unsupervisedly, and henceforth, a general formulation for neural cloth simulation. The key to achieve this is to adapt an existing optimization scheme for motion from simulation based methodologies to deep learning. Then, analyzing the nature of the problem, we devise an architecture able to automatically disentangle static and dynamic cloth subspaces by design. We will show how this improves model performance. Additionally, this opens the possibility of a novel motion augmentation technique that greatly improves generalization. Finally, we show it also allows to control the level of motion in the predictions. This is a useful, never seen before, tool for artists. We provide of detailed analysis of the problem to establish the bases of neural cloth simulation and guide future research into the specifics of this domain.
How to read this
- Category
- Method: unsupervised neural garment dynamics
- Contributions
- First unsupervised deep-learning framework for realistic cloth dynamics, using physics-inspired losses with no ground-truth simulation data
- An architecture that by design disentangles static and dynamic cloth subspaces, with a motion-augmentation technique that improves generalization
- Provides artist control over the level of motion in predictions
- Context
- Extends the unsupervised physically-based-neural-simulation line, notably the authors' PBNS (2021), from static pose-space deformation to true cloth dynamics.Builds on: PBNS: Physically Based Neural Simulation for Unsupervised Garment Pose Space Deformation
- Correctness
- Trained without ground-truth data via physics-inspired losses, so realism depends on the loss formulation rather than reference simulation; the paper positions itself as establishing foundations, so treat results as a general formulation more than an exhaustively benchmarked solver.
- Clarity
- Analytical and reasonably accessible; a first pass conveys the unsupervised dynamics idea, a second pass covers the static/dynamic disentanglement and the optimization-from-simulation adaptation.
- How to read it
- First pass for the unsupervised dynamics premise and the static/dynamic split; second pass on the physics-inspired losses and motion augmentation if you intend to build on PBNS-style training.
Builds on
Related work
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →