MULTI-LAYERED 3D GARMENTS ANIMATION

ABSTRACT

Most existing 3D garment animation datasets are restricted to human bodies with single-layered garments. Even though cases with upper shirts and lower pants are included, only a few overlap areas among such garment combinations exist. Moreover, they often regard human body movement as the only driving factor that causes garment animation. Approaches developed on top of these datasets thus tend to model garments as functions of human body parameters such as body shape and pose. While such treatment leads to promising performance on existing datasets, it leaves a gap between experimental environments and real scenarios, where a body can wear multiple layered garments and the corresponding garment dynamics can be affected by environmental factors and garment attributes. Consequently, existing approaches often struggle to generalize to multi-layered garments and realistic scenarios. To facilitate the advance of 3D garment animation toward handling more challenging cases, this paper presents a new large-scale synthetic dataset called LAYERS, covering 4,900 different combinations of multi-layered garments with 700k frames in total. The animation of these multi-layered garments follows the laws of physics and is affected by not only human body movements but also random environmental wind and garment attributes. To demonstrate the quality of LAYERS, we further propose a novel method, LayersNet, for 3D garment animation, which represents garments as unions of particles and subsequently adopts a neural network to animate garments via particle-based simulation. In this way, the interactions between different parts of one garment, different garments on the same body, and garments against various driving factors, can be naturally and uniformly handled via the interactions of particles. Through comprehensive experiments, LayersNet demonstrates superior performance in terms of animation accuracy and generality over baselines. The proposed dataset, LAY-ERS, as well as the proposed method, LayersNet, will be publicly available.

1. INTRODUCTION

3D garment animation has been an active and important topic in computer graphics and machine learning, due to its great potential in various downstream tasks, including virtual reality, virtual tryon, gaming and film production. While this topic has been extensively studied in the past, generating realistic and faithful animation remains an open research question. In particular, existing approaches are still limited in modeling diverse garments of different topologies and appearances. In addition, the complex interactions between the garment and the human body under the challenging setting of multi-layer garments and with external environmental factors remain much less explored in the literature.



Figure 1: We propose a new large-scale 3D garment animation dataset LAYERS, which improves over previous datasets by considering multi-layered 3D garments and more driving factors for garment animation, e.g., environmental wind, besides human body movements. In (a)-(d) we show the new and realistic challenges covered in LAYERS but are omitted in previous datasets.

