← ArchivePaper2021
PBNS: Physically Based Neural Simulation for Unsupervised Garment Pose Space Deformation
Formulates physics-based simulation as an implicit deep learning loss to unsupervisedly learn garment pose space deformation bases for dressed humans, matching PBS quality in comparable training time.
Abstract
We present a methodology to automatically obtain Pose Space Deformation (PSD) basis for rigged garments through deep learning. Classical approaches rely on Physically Based Simulations (PBS) to animate clothes. These are general solutions that, given a sufficiently fine-grained discretization of space and time, can achieve highly realistic results. However, they are computationally expensive and any scene modification prompts the need of re-simulation. Linear Blend Skinning (LBS) with PSD offers a lightweight alternative to PBS, though, it needs huge volumes of data to learn proper PSD. We propose using deep learning, formulated as an implicit PBS, to unsupervisedly learn realistic cloth Pose Space Deformations in a constrained scenario: dressed humans. Furthermore, we show it is possible to train these models in an amount of time comparable to a PBS of a few sequences. To the best of our knowledge, we are the first to propose a neural simulator for cloth. While deep-based approaches in the domain are becoming a trend, these are data-hungry models. Moreover, authors often propose complex formulations to better learn wrinkles from PBS data. Supervised learning leads to physically inconsistent predictions that require collision solving to be used.
How to read this
- Category
- Method: a neural cloth simulator (unsupervised garment PSD learning)
- Contributions
- Formulates physically based simulation as an implicit deep learning loss to learn garment Pose Space Deformation bases without ground-truth simulation data
- Trains a neural cloth model for dressed humans in time comparable to running PBS on a few sequences
- Presents what the authors describe as the first neural simulator for cloth
- Context
- Targets the same problem as learning-based virtual try-on (Santesteban et al.), replacing data-hungry supervised PSD learning with a self-supervised, physics-loss formulation over LBS-rigged garments.Builds on: Learning-Based Animation of Clothing for Virtual Try-On
- Correctness
- Validated in a constrained scenario (dressed humans on a parametric body), and being unsupervised it sidesteps PBS data collection, but the constrained setting and the chosen physics terms bound how general or accurate the learned deformations are.
- Clarity
- Accessible at the idea level; a first pass conveys the unsupervised-physics-loss concept, a second pass is needed for the loss formulation and PSD parameterization.
- How to read it
- First pass for the implicit-PBS-as-loss idea and why it avoids simulation data; do a second pass on the loss terms and the LBS plus PSD setup if you intend to reimplement or compare against PBS.
Builds on
Built upon by
Related work
- Neural Cloth Simulation 2022 / SIGGRAPH Asia
- SNUG: Self-Supervised Neural Dynamic Garments 2022 / CVPR
- SwinGar: Spectrum-Inspired Neural Dynamic Deformation for Free-Swinging Garments 2024 / TVCG
- Motion Guided Deep Dynamic 3D Garments 2022 / SIGGRAPH Asia
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 →