Neural Network Garment Collision Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for simulating three-dimensional garment deformations and handling garment-character body collisions are computationally intensive and often produce inaccurate results, failing to provide real-time performance for interactive applications like gaming and virtual try-on.
Innovation Solution
A digital design system utilizing three neural networks to predict garment deformations and resolve collisions in real-time by identifying penetrating vertices and calculating offset distances to move them outside the character body, using a Repulsive Force Unit (ReFU) to model repulsive forces and preserve geometric details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If physically based simulation methods (mass-spring system, finite element approach) are used to generate accurate cloth deformations, then manufacturing precision is improved, but productivity deteriorates due to computational intensity
Solution Approach 1:
The patent replaces traditional mechanically-based physics simulation systems (mass-spring, finite element) with a neural network-based machine learning system. The neural network is trained offline on physics simulation data and then deployed for real-time inference, substituting the computational mechanics approach with a learned predictive model that delivers both accuracy and speed.
Solution Approach 2:
The patent performs comprehensive garment deformation simulations using physically-based models during the offline training phase to generate labeled training data. This preliminary action creates a dataset that captures accurate physics-based deformations, which are then used to train the neural network for efficient real-time prediction without requiring repeated expensive physics simulations.
2Productivity
If machine learning methods are used to predict three-dimensional clothing deformations, then productivity is improved, but manufacturing precision deteriorates due to inaccurate results
Solution Approach 1:
The patent uses neural networks to replace traditional mechanics-based physics solvers for real-time garment deformation prediction. The network learns the complex nonlinear relationships between character pose and garment deformation from training data, enabling fast inference that maintains physical accuracy without requiring iterative mechanical simulations during runtime.
Solution Approach 2:
The patent transforms the physics simulation problem into a machine learning parameter prediction problem. Instead of solving differential equations and constraint systems in real-time, the neural network directly predicts garment vertex positions as a function of character body parameters and pose parameters, changing the computational approach from continuous physics solving to discrete parameter mapping.
3Manufacturing precision
If existing methods are used for collision detection and resolution, then manufacturing precision is improved, but productivity deteriorates due to computational overhead
Solution Approach 1:
The patent replaces traditional geometric collision detection and resolution algorithms with a neural network-based approach. The network is trained to predict garment deformations that inherently avoid penetration into the character body, eliminating the need for separate collision detection and iterative resolution steps required by conventional methods.
Solution Approach 2:
The patent merges the garment deformation prediction and collision avoidance functions into a single unified neural network model. Rather than separately computing deformations and then resolving collisions, the network simultaneously predicts deformations that naturally respect collision constraints, combining multiple functions into one efficient computational process.
Data Source
AI summary
Embodiments are disclosed for using machine learning models to perform three-dimensional garment deformation due to character body motion with collision handling. In particular, in one or more embodiments, the disclosed systems and methods comprise receiving an input, the input including character body shape parameters and character body pose parameters defining a character body, and garment parameters. The disclosed systems and methods further comprise generating, by a first neural network, a first set of garment vertices defining deformations of a garment with the character body based on the input. The disclosed systems and methods further comprise determining, by a second neural network, that the first set of garment vertices includes a second set of garment vertices penetrating the character body. The disclosed systems and methods further comprise modifying, by a third neural network, each garment vertex in the second set of garment vertices to positions outside the character body.