Towards Multi-Layered 3D Garments Animation

ICCV, 2023

Realistic Predictions (Non-SMPL Human)

D-LAYERS Dataset Samples



Mimicking realistic dynamics in 3D garment animations is a challenging task due to the complex nature of multi-layered garments and the variety of outer forces involved. Existing approaches mostly focus on single-layered garments driven by only human bodies and struggle to handle general scenarios. In this paper, we propose a novel data-driven method, called LayersNet, to model garment-level animations as particle-wise interactions in a micro physics system. We improve simulation efficiency by representing garments as patch-level particles in a two-level structural hierarchy. Moreover, we introduce a novel Rotation Equivalent Transformation with Rotation Invariant Attention that leverage the rotation invariance and additivity of physics systems to better model outer forces. To verify the effectiveness of our approach and bridge the gap between experimental environments and real-world scenarios, we introduce a new challenging dataset, D-LAYERS, containing 700K frames of dynamics of 4,900 combinations of multi-layered garments driven by human bodies and randomly sampled wind. Our LayersNet achieves superior performance both quantitatively and qualitatively.



Diverse Dynamics

Triple Garments


Qualitative Results

Predicted Animatioins with Customized Settings

Predicted Animatioins

Baseline Comparisons



Euclidean error (mm) on sampled D-LAYERS with maximum sequence length of 35 frames. The collision rates between different layers of garments are shown under L-Collision, while the collision rates between garments and human bodies are shown under H-Collision. Models trained with full collision loss are marked by +. Our LayersNet achieves superior results in all cases.

Methods Jacket Jacket + Hood Dress Jumpsuit Skirt
DeePSD+ 1830.1±803.3 1566.0±527.1 1333.0±349.2 1219.0±186.8 1194.7±311.2
GarSim+ 1412.1±886.8 1139.1±653.5 674.4±451.8 317.8±157.4 689.9±386.7
LayersNet(Ours) 571.9±451.9 493.9±354.2 397.2±342.2 264.0±200.2 301.3±79.3
LayersNet+(Ours) 567.3±425.5 491.4±361.3 379.1±299.7 260.1±222.2 299.5±92.3
Methods Pants T-shirt Overall L-Collision(%) H-Collision(%)
DeePSD+ 1185.7±213.3 1202.9±233.6 1563.4.9±486.8 8.78±5.12 19.47±6.38
GarSim+ 317.8±150.1 447.6±303.8 1028.3±581.0 6.03±4.23 15.11±7.11
LayersNet(Ours) 234.4±206.3 273.3±169.0 472.8±343.5 3.13±2.22 10.68±4.53
LayersNet+(Ours) 200.9±140.1 267.8±189.6 467.2±330.7 3.77±2.60 2.16±1.46



 author = {Shao, Yidi and Loy, Chen Change and Dai, Bo},
 title = {Towards Multi-Layered 3D Garments Animation},
 booktitle = {ICCV},
 year = {2023}


Shao Yidi
Email: yidi001 at