Parametric Linear Blend Skinning Model for Multiple-Shape 3D Garments.

Area of Science:

• Computer Graphics
• Animation
• Machine Learning

Background:

• Existing data-driven 3D garment animation methods often rely heavily on human body models.
• These methods exhibit limited adaptability to diverse garment shapes and sizes.
• Challenges include over-reliance on human body topology and difficulty generalizing across different clothing types.

Purpose of the Study:

• To develop a novel data-driven Parametric Linear Blend Skinning (PLBS) model for generalized 3D garment dressing and animation.
• To overcome the limitations of previous methods by shifting focus from human body modeling to garment modeling.
• To enable adaptable and generalized modeling of various garments and body types.

Main Methods:

• Formulated garment deformation using a PLBS model controlled by a canonical 3D garment mesh, vertex-based skinning weights, and local patch transformations.
• Developed a garment-centric modeling approach, separating in-plane (shape/size variation) and out-of-plane (body/motion deformation) subspaces.
• Introduced novel 3D garment registration and skinning weight decomposition strategies for data acquisition.
• Employed dynamic fine-tuning to enhance high-frequency signal representation for unseen data.

Main Results:

• The PLBS model effectively models dynamics for loose-fitting garments, outperforming prior data-driven methods.
• Demonstrated the model's ability to factorize and generalize across varied body sizes, garment shapes, garment sizes, and human motions.
• Achieved uniform expression of varied garments and bodies through distinct in-plane and out-of-plane deformation controls.

Conclusions:

• The proposed PLBS model offers a robust and generalized solution for 3D garment animation, surpassing previous data-driven techniques.
• This garment-centric approach enhances adaptability and scalability for diverse clothing and body types.
• The method successfully addresses key challenges in 3D garment modeling and animation, paving the way for more realistic digital clothing.