AgentDress: Realtime Clothing Synthesis for Virtual Agents using Plausible Deformations

Area of Science:

• Computer Graphics
• Virtual Reality
• Animation

Background:

• Real-time animation of virtual humans requires efficient methods for simulating clothing deformation.
• Existing methods often struggle to accurately represent the complex interplay between body shape, pose, and cloth dynamics.

Purpose of the Study:

• To develop a computationally efficient, CPU-based method for real-time cloth animation that accounts for variations in virtual human shape and pose.
• To improve the visual fidelity and user perception of dressed virtual agents in immersive virtual environments.

Main Methods:

• Formulated clothing deformation as a function of body shape and pose parameters, factorized into independent Clothing Pose and Shape Models.
• Employed pose clustering and sensitivity-based distance measurements to efficiently calculate and blend Taylor expansion results for animation.
• Achieved real-time performance (50+ FPS) on CPU for clothing with tens of thousands of vertices.

Main Results:

• The proposed method achieves real-time cloth animation on a CPU, outperforming traditional linear blend skinning in user studies.
• Successfully integrated shape variation into existing clothing pose models, offering greater generality.
• Demonstrated improved user perception of dressed virtual agents in immersive virtual environments.

Conclusions:

• The presented approach offers a significant advancement in real-time cloth animation for virtual humans, balancing performance and visual quality.
• This method enhances the immersion and realism of virtual characters by accurately simulating clothing responses to body dynamics.
• The factorization and blending techniques provide a scalable and adaptable solution for various virtual character applications.