DeepCloth: Neural Garment Representation for Shape and Style Editing

Area of Science:

• Computer Vision and Graphics
• 3D Digitalization

Background:

• Existing garment representations struggle with smooth transitions across shapes and topologies.
• 3D garment digitization, editing, and animation remain challenging research areas.

Purpose of the Study:

• Introduce DeepCloth, a unified framework for garment representation, reconstruction, animation, and editing.
• Address limitations in current methods for handling diverse garment geometries and dynamic changes.

Main Methods:

• Utilize a topology-aware UV-position map and mask for unified garment geometry description.
• Embed UV-based representations into a continuous feature space for latent space optimization and control.
• Unify neural garment representation with body shape and pose for plausible animation.

Main Results:

• Achieve state-of-the-art performance in garment representation compared to prior methods.
• Demonstrate plausible garment animation results, even with significant editing operations.
• Enable flexible and general 3D garment digitization.

Conclusions:

• DeepCloth provides a significant advancement in 3D garment digitization.
• The framework offers a unified approach to garment representation, reconstruction, and animation.
• The method shows potential for broader applications in digital fashion and virtual reality.