Garment Draping over the Human Body
Event details
| Date | 01.07.2022 |
| Hour | 09:30 › 11:30 |
| Speaker | Ren LI |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Sabine Süsstrunk
Thesis advisor: Prof. Pascal Fua
Co-examiner: Prof. Ronan Boulic
Abstract
We consider the problem of draping garment over posed human body. Traditionally, the physics-based simulator is used to model garment deformation due to the high fidelity of simulation results. However, its computation cost is notoriously heavy, which hinders its deployment in real-world applications. To meet the real-time or near-real time requirement, recent works propose learning-based methods to learn how to deform garment in supervised or self-supervised fashions. Although these methods are able to produce plausible dynamics and details with fast speed, we observe that none of them is fully differentiable and scalable to garment with any topology. This observation leads to our proposed method that is end-to-end differentiable, capable to capture fine details of garment in arbitrary topology, and even supports topology changing according to external signals, e.g., 3D scans, 2D segmentation or sketches of garment.
Background papers
Exam president: Prof. Sabine Süsstrunk
Thesis advisor: Prof. Pascal Fua
Co-examiner: Prof. Ronan Boulic
Abstract
We consider the problem of draping garment over posed human body. Traditionally, the physics-based simulator is used to model garment deformation due to the high fidelity of simulation results. However, its computation cost is notoriously heavy, which hinders its deployment in real-world applications. To meet the real-time or near-real time requirement, recent works propose learning-based methods to learn how to deform garment in supervised or self-supervised fashions. Although these methods are able to produce plausible dynamics and details with fast speed, we observe that none of them is fully differentiable and scalable to garment with any topology. This observation leads to our proposed method that is end-to-end differentiable, capable to capture fine details of garment in arbitrary topology, and even supports topology changing according to external signals, e.g., 3D scans, 2D segmentation or sketches of garment.
Background papers
- Narain et al. Adaptive Anisotropic Remeshing for Cloth Simulation - https://dl.acm.org/doi/pdf/10.1145/2366145.2366171
- Santesteban et al. SNUG: Self-Supervised Neural Dynamic Garments - http://mslab.es/projects/SNUG/contents/santesteban_CVPR2022.pdf
- Corona et al. SMPLicit: Topology-aware Generative Model for Clothed People - https://openaccess.thecvf.com/content/CVPR2021/papers/Corona_SMPLicit_Topology-Aware_Generative_Model_for_Clothed_People_CVPR_2021_paper.pdf
Practical information
- General public
- Free