Extended Short- and Long-Range Mesh Learning for Fast and Generalized Garment Simulation

TL;DR

This work tackles the computational bottleneck of graph neural networks in high-resolution garment simulation by introducing two modules: Laplacian Smoothed Dual Message-Passing (LSDMP) for extended short-range propagation and Geodesic Self-Attention (GSA) for long-range, geodesic-aware interactions. The two modules run in parallel within a unified pipeline, enabling effective modeling of both local elasticity and global mesh structure without resorting to multi-scale meshes. Empirical results on AMASS-based data show state-of-the-art physical fidelity with fewer layers and lower inference latency, and strong generalization to unseen garments and avatars. The approach advances real-time, pose-conditioned garment simulation by delivering accurate dynamics with improved efficiency and scalability.

Abstract

3D garment simulation is a critical component for producing cloth-based graphics. Recent advancements in graph neural networks (GNNs) offer a promising approach for efficient garment simulation. However, GNNs require extensive message-passing to propagate information such as physical forces and maintain contact awareness across the entire garment mesh, which becomes computationally inefficient at higher resolutions. To address this, we devise a novel GNN-based mesh learning framework with two key components to extend the message-passing range with minimal overhead, namely the Laplacian-Smoothed Dual Message-Passing (LSDMP) and the Geodesic Self-Attention (GSA) modules. LSDMP enhances message-passing with a Laplacian features smoothing process, which efficiently propagates the impact of each vertex to nearby vertices. Concurrently, GSA introduces geodesic distance embeddings to represent the spatial relationship between vertices and utilises attention mechanisms to capture global mesh information. The two modules operate in parallel to ensure both short- and long-range mesh modelling. Extensive experiments demonstrate the state-of-the-art performance of our method, requiring fewer layers and lower inference latency.

Figures (7)

• Figure 1: Laplacian smoothing-based propagation allows features to reach distant vertices much earlier in an attenuated fashion, and is vastly more efficient than conventional message-passing. Geodesic self-attention further enables graph-aware long-range connections without message-passing mechanisms.
• Figure 2: Overview of the proposed method for 3D garment simulation. It consists of two novel modules: LSDMP (Laplacian Smoothed Dual Message-Passing) and GSA (Geodesic Self-Attention), to process extended short and long-range mesh learning, respectively.
• Figure 3: LSDMP extends conventional message-passing using Laplacian-smoothing to efficiently and emblematically propagate vertex features.
• Figure 4: GSA injects geodesic information into vertex features by appending MDS-reduced coordinates derived from the geodesic distance matrix.
• Figure 5: Qualitative comparison with SOTA methods. We further include a physical simulation method - ARCSim narain2012adaptive for reference.