Capturing and Modeling Real Cloth Deformations for Virtual Garment Design

Abstract

Accurate simulation of cloth deformations is a very challenging task that is mainly addressed using pure physically based techniques. Data-driven approaches have been recently proposed using 3D scanning of real-dressed humans. However, the noise and irregularities in the acquired 3D meshes require intense work for data cleaning and handling. Moreover, the acquired data need additional properties to be animated such as skinning weight and fabric parameters. In this paper, we present a comprehensive pipeline that integrates realistic geometries from real-world scans with desired properties to create animation-ready 3D models. Our approach robustly matches and transfers necessary animation information from an artist-defined 3D model to a 3D scan of a clothed human. This results in a fully automatic process that transforms a single 3D scan into an animated 3D setup for modeling and simulation of clothed characters. In this way, the output of our method can provide a dataset of accurate cloth deformations for the training of new generative cloth simulation systems. We envisage our work in the fashion industry for garment design to improve the evaluation of the dynamic cloth behavior before the physical production. This aspect could speed up the design process and at the same time limit the use of physical fabrics reducing costs and environmental impact.