Papers
Topics
Authors
Recent
Search
2000 character limit reached

Predicting Loose-Fitting Garment Deformations Using Bone-Driven Motion Networks

Published 3 May 2022 in cs.GR, cs.CV, and cs.LG | (2205.01355v3)

Abstract: We present a learning algorithm that uses bone-driven motion networks to predict the deformation of loose-fitting garment meshes at interactive rates. Given a garment, we generate a simulation database and extract virtual bones from simulated mesh sequences using skin decomposition. At runtime, we separately compute low- and high-frequency deformations in a sequential manner. The low-frequency deformations are predicted by transferring body motions to virtual bones' motions, and the high-frequency deformations are estimated leveraging the global information of virtual bones' motions and local information extracted from low-frequency meshes. In addition, our method can estimate garment deformations caused by variations of the simulation parameters (e.g., fabric's bending stiffness) using an RBF kernel ensembling trained networks for different sets of simulation parameters. Through extensive comparisons, we show that our method outperforms state-of-the-art methods in terms of prediction accuracy of mesh deformations by about 20% in RMSE and 10% in Hausdorff distance and STED. The code and data are available at https://github.com/non-void/VirtualBones.

Citations (41)

Summary

  • The paper introduces a novel two-stage algorithm that first predicts low-frequency deformations using GRU-driven virtual bones and then refines details via a GNN.
  • The method achieves roughly 20% improvement in RMSE and 10% gains in Hausdorff distance and STED compared to state-of-the-art models.
  • This approach enables realistic and interactive simulation of garment dynamics, benefiting applications in gaming and virtual reality.

Predicting Loose-Fitting Garment Deformations Using Bone-Driven Motion Networks

The paper, titled "Predicting Loose-Fitting Garment Deformations Using Bone-Driven Motion Networks," addresses the challenges in accurately predicting the deformation of loose-fitting garments, which exhibit complex dynamic behaviors not closely following human body shapes. This research contributes a novel approach different from existing learning-based methods that predominantly focus on tight-fitting garments or static poses. It introduces a two-step learning algorithm integrated with bone-driven motion networks to efficiently predict garment deformations at interactive rates.

Methodology Overview

The core methodology involves a two-stage process to predict low- and high-frequency garment deformations using virtual bones derived from pre-computed simulation data. This approach effectively divides the complex task of garment deformation into two manageable sub-tasks:

  1. Low-Frequency Deformation Prediction: The first stage involves transforming body motions into the motions of virtual bones using Gated Recurrent Units (GRU). These virtual bones, extracted using skinning decomposition techniques like SSDR, represent the overall shape and basic motion of the garment.
  2. High-Frequency Deformation Estimation: The second stage uses the outputs from the low-frequency predictions and processes them with local features extracted from the garment mesh through a Graph Neural Network (GNN). By combining global information of bone motions and mesh-local details, the method enhances the garment's surface detail predictions, including dynamic folds and wrinkles.

To handle changes in simulation parameters, the researchers employ a Radial Basis Function (RBF) kernel to interpolate results from multiple trained networks, each corresponding to a specific set of parameters. This enables the network to generalize deformations under varied simulation conditions without explicit retraining for each new parameter set.

Experimental Results

The researchers validate their approach using a dataset composed of garments simulated with various parameters, demonstrating the capabilities of the model under different body motions and simulation settings. Impressively, the proposed method improves prediction accuracy by approximately 20% in Root Mean Squared Error (RMSE) and 10% in Hausdorff distance and Spatio-Temporal Edge Difference (STED) over comparable state-of-the-art methods, confirming its robustness and accuracy in handling non-trivial garment dynamics.

Implications and Future Directions

The implications of this work are multifaceted. Practically, the proposed method offers a significant improvement in interactive applications such as gaming and virtual reality, where realistic garment simulation at real-time speeds is crucial. Theoretically, the introduction of virtual bones as predictive intermediaries presents an innovative way to cater to garments' intricate deformations and could be adapted for various other deformable object simulations.

Looking forward, the research could extend into optimized bone extraction algorithms to further improve the low-frequency module's accuracy. Additionally, incorporating collision detection mechanisms directly into the high-frequency predictor may enhance garment-body interaction accuracy, handling self-collision artifacts more effectively.

This research effectively leverages cutting-edge machine learning techniques to address a long-standing challenge in the domain of virtual garment simulation, offering a promising path for future advancements in related areas.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.

Collections

Sign up for free to add this paper to one or more collections.

GitHub