Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GGAvatar: Reconstructing Garment-Separated 3D Gaussian Splatting Avatars from Monocular Video

Jingxuan Chen

TL;DR

GGAvatar (Garment-separated 3D Gaussian Splatting Avatar), which relies on monocular videos, effectively achieves decoupled, editable, and realistic reconstruction of clothed humans.

Abstract

Avatar modelling has broad applications in human animation and virtual try-ons. Recent advancements in this field have focused on high-quality and comprehensive human reconstruction but often overlook the separation of clothing from the body. To bridge this gap, this paper introduces GGAvatar (Garment-separated 3D Gaussian Splatting Avatar), which relies on monocular videos. Through advanced parameterized templates and unique phased training, this model effectively achieves decoupled, editable, and realistic reconstruction of clothed humans. Comparative evaluations with other costly models confirm GGAvatar's superior quality and efficiency in modelling both clothed humans and separable garments. The paper also showcases applications in clothing editing, as illustrated in Figure 1, highlighting the model's benefits and the advantages of effective disentanglement. The code is available at https://github.com/J-X-Chen/GGAvatar/.

GGAvatar: Reconstructing Garment-Separated 3D Gaussian Splatting Avatars from Monocular Video

TL;DR

GGAvatar (Garment-separated 3D Gaussian Splatting Avatar), which relies on monocular videos, effectively achieves decoupled, editable, and realistic reconstruction of clothed humans.

Abstract

Avatar modelling has broad applications in human animation and virtual try-ons. Recent advancements in this field have focused on high-quality and comprehensive human reconstruction but often overlook the separation of clothing from the body. To bridge this gap, this paper introduces GGAvatar (Garment-separated 3D Gaussian Splatting Avatar), which relies on monocular videos. Through advanced parameterized templates and unique phased training, this model effectively achieves decoupled, editable, and realistic reconstruction of clothed humans. Comparative evaluations with other costly models confirm GGAvatar's superior quality and efficiency in modelling both clothed humans and separable garments. The paper also showcases applications in clothing editing, as illustrated in Figure 1, highlighting the model's benefits and the advantages of effective disentanglement. The code is available at https://github.com/J-X-Chen/GGAvatar/.

Paper Structure

This paper contains 14 sections, 10 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Parametric Representation
  4. Method
  5. Garment Templates Estimation
  6. Gaussian Representation and Deformation
  7. Rendering with Splatting.
  8. Training Losses
  9. Experiments
  10. Dataset
  11. Comparison with SoTA Methods
  12. Ablation Study
  13. Application
  14. Conclusion

Figures (5)

  • Figure 1: The Framework of GGAvatar. The model accepts monocular RGB images or videos along with their masks as input, producing an editable clothed human avatar that can be viewed from any perspective for any pose. The red line symbolizes the garment template modelling procedure, the orange line represents the deformation process, and the blue line indicates two types of rendering optimization.
  • Figure 2: Results of novel view synthesis on the ZJU-MoCap.
  • Figure 3: The results of holistic clothing transfer between GGAvatar and SCARFscarf (under two different training times). The modelling garments are displayed by attaching them to another person on the People Snapshot datasetpeoplesnapshot.
  • Figure 4: Ablation results for the clothes from male-3-causal are obtained. Due to consistent Gaussian rendering methods employed, the middle regions in the figure showcase similar reconstruction effects.
  • Figure 5: The clothing transfer and the colour-changing application of GGAvatar.