FlexPara: Flexible Neural Surface Parameterization.

Area of Science:

• Computer Graphics
• Computational Geometry
• Geometric Deep Learning

Background:

• Surface parameterization is crucial for 3D asset visualization and analysis.
• Existing methods require high-quality meshes and manual intervention for complex topologies.
• Adaptable parameterization pipelines are needed for diverse surface structures and tasks.

Purpose of the Study:

• Introduce FlexPara, an unsupervised neural optimization framework for flexible surface parameterization.
• Enable both global and multi-chart parameterizations without manual seam specification.
• Provide a controllable processing pipeline for geometry processing tasks.

Main Methods:

• Developed a bi-directional cycle mapping framework using geometrically-interpretable sub-networks.
• Implemented functionalities for cutting, deforming, unwrapping, and wrapping 3D surfaces.
• Constructed an adaptive chart assignment mechanism for multi-chart parameterization.

Main Results:

• Achieved automated global and multi-chart surface parameterization.
• Demonstrated superior performance compared to conventional methods.
• Showcased the framework's universality and potential across various applications.

Conclusions:

• FlexPara offers a novel, unsupervised approach to surface parameterization.
• The framework provides flexibility and control, overcoming limitations of traditional methods.
• The neural paradigm presents significant potential for future geometry processing advancements.