Template-based representation sharing the same texture information which achieves a consistent texture for the composed clothing animation. (a) Garment template with texture information. (b) Texture-consistent garment animation.

Processing math: 100%

Download： Full Size Img PowerPoint

Template-based representation sharing the same texture information which achieves a consistent texture for the composed clothing animation. (a) Garment template with texture information. (b) Texture-consistent garment animation.

2023, 38(6): 1356-1368.

Motion-Inspired Real-Time Garment Synthesis with Temporal-Consistency .

Figures of the Article

Learning-based garment animation workflow translating the input body motion into clothing dynamics. The efficient workflow is spatially and temporally consistent, easy to control, and capable of garments that are different from body topology.
Overview of our method. (a) Data generation. (b) Garment synthesis network. (c) Runtime stage.
Detailed architecture of our garment synthesis network as discussed in Subsection 3.3.
We employ a refinement procedure to remove the penetration between the garment and the body. (a) Penetrations between the data-driven clothing deformations and the underlying body. (b) Results from our collision removal process. (c) Collision removal results from [5]. Details are enlarged to show the difference.
Template-based representation sharing the same texture information which achieves a consistent texture for the composed clothing animation. (a) Garment template with texture information. (b) Texture-consistent garment animation.
Garment template used in our experiment. (a) Garment pattern connected by sewing lines. (b) Simulated garment under the body mesh.
Evaluation of hyper-parameters of our network with RMSE (root mean squared error) as the reference. Different settings result in different reconstruction errors. We select different (a) $seq\_len$ and (b) $emb\_size$ that is represented by 2ⁿ to train our network. Comparatively, the network performs best when $seq\_len = 6$ and $emb\_size = 2^{8} = 256$ .
Reconstruction performance using different methods. (a) Ground-truth. (b) Ours. (c) Wang et al.^[2]. (d) RNN^[9]. (e) GRU^[7]. (f) LSTM^[10]. Details are enlarged to show the difference.
Visualization of reconstruction errors utilizing a difference map. The predicted garments are encoded as hot/cold colors, representing large/small per-vertex errors. (a) Ground truth (simulated garments). (b)–(f) Prediction by our method, Wang et al.^[2], RNN^[9], GRU^[7], and LSTM^[10] respectively.
Garment animation synthesized by our workflow on a motion sequence not used for training. It infers how the garment dynamics deform with the body motion.
Synthesized garments on unseen poses.

Others

External link to attachment
https://rdcu.be/dxI8E
PDF format
2023-6-10-1887-Highlights 249KB
Compressed file
2023-6-10-1887-Highlights 125KB

Related articles

CTNet: A Convolutional Transformer Network for Color Image Steganalysis

2025, 40(2): 413-427. DOI: 10.1007/s11390-023-3006-3
Learning a Mixture of Conditional Gating Blocks for Visual Question Answering

2024, 39(4): 912-928. DOI: 10.1007/s11390-024-2113-0
OAAFormer: Robust and Efficient Point Cloud Registration Through Overlapping-Aware Attention in Transformer

2024, 39(4): 755-770. DOI: 10.1007/s11390-024-4165-6
CAT: A Simple yet Effective Cross-Attention Transformer for One-Shot Object Detection

2024, 39(2): 460-471. DOI: 10.1007/s11390-024-1743-6
WavEnhancer: Unifying Wavelet and Transformer for Image Enhancement

2024, 39(2): 336-345. DOI: 10.1007/s11390-024-3414-z
A Comparative Study of CNN- and Transformer-Based Visual Style Transfer

2022, 37(3): 601-614. DOI: 10.1007/s11390-022-2140-7
Temporally Consistent Depth Map Prediction Using Deep CNN and Spatial-temporal Conditional Random Field

2017, 32(3): 443-456. DOI: 10.1007/s11390-017-1735-x
On the Arc Consistency Problem

1999, 14(4): 298-308.
A Framework of Memory Consistency Models

1998, 13(2): 110-124.
Denote CSP with Temporal Formulas

1990, 5(1): 17-23.