DiffAvatar: Simulation-Ready Garment Optimization with Differentiable Simulation
Abstract
The realism of digital avatars is crucial in enabling telepresence applications with self-expression and customization. A key aspect of this realism originates from the physical accuracy of both a true-to-life body shape and clothing. While physical simulations can produce high-quality, realistic motions for clothed humans, they require precise estimation of body shape and high-quality garment assets with associated physical parameters for cloth simulations. However, manually creating these assets and calibrating their parameters is labor-intensive and requires specialized expertise. To address this gap, we propose DiffAvatar, a novel approach that performs body and garment co-optimization using differentiable simulation. By integrating physical simulation into the optimization loop and accounting for the complex nonlinear behavior of cloth and its intricate interaction with the body, our framework recovers body and garment geometry and extracts important material parameters in a physically plausible way. Our experiments demonstrate that our approach generates realistic clothing and body shape that can be easily used in downstream applications.
Pipeline
DiffAvatar generates simulation-ready avatar assets from inputs obtained through a multi-view capture. Our pipeline initially preprocesses the 3D scan to segment the target garment and establish the initial pose and shape of the parametric body model. We employ a differentiable simulation framework to align our simulated garment with the segmented garment by jointly optimizing the garment’s design and material parameters
Garment 2D Representation
Garments can take on a wide range of 3D shapes when draped onto a body, due to factors such as changing pose and dynamics or wearer manipulations. Despite this large variation in configurations, garments are compactly represented by their 2D patterns (Fig. 3), which consist of the individual pieces of fabric that are sewn together to create the 3D clothing. Therefore, we represent clothing in 2D pattern space, which ensures developable meshes and manufacturable clothing. Virtual garments are modeled as triangle meshes, with their rest shape encoded in these 2D patterns. The rest shape is crucial for modeling the in-plane stretching and shearing behavior of different fabrics.
Control Cage Garment Shape Optimization
We propose a regularized differentiable cage formulation to effectively and robustly optimize for the 2D patterns of garments such that the simulated and draped 3D representation of the garment closely aligns with the scan.
Control Cage Pattern Representation:
While it is possible to directly optimize for the 2D pattern vertices p directly, this approach is highly non-regularized and can produce ill-shaped or even non-physical inverted rest shape geometries that cause simulators to fail.
A high number of optimization variables can also cause the optimization to get stuck in a local minimum (See our ablation study in Sec. 5.4). Additionally, directly optimizing for the 2D coordinates does not respect design constraints that are better represented in a limited subspace of reasonable designs. Therefore, we further regulate the optimization problem by selecting and optimizing a set of 2D control vertices ζ on the boundaries of the individual panels of the 2D pattern that directly deform and manipulate the underlying 2D patterns through control cages instead.
Control Cage Handle Selection: We use the geometric information of the 2D garment patterns to automatically identify control cage points, see the inset figure above. Our algorithm first extracts the boundary loop of the underlying mesh for each connected component representing a garment panel in the 2D garment pattern, then processes the boundary loop and marks a vertex as a control point if it lies on the convex hull of the pattern or when its local curvature exceeds a threshold (10° in our implementation).
Demo
Citation
@inproceedings{li2023diffavatar,
title={{DiffAvatar}: Simulation-Ready Garment Optimization with Differentiable Simulation},
author={Li, Yifei and Chen, Hsiao-yu and Larionov, Egor and Sarafianos, Nikolaos and Matusik, Wojciech and Stuyck, Tuur},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
eprint={2311.12194},
year = {2024}
}
Keywords
graphics, physical simulation, digital avatar, differentiable simulation, cloth simulation