SketchMetaFace: A Learning-Based Sketching Interface for High-Fidelity 3D Character Face Modeling.

Area of Science:

• Computer Graphics
• Human-Computer Interaction
• Artificial Intelligence

Background:

• 3D avatar creation is crucial for AR/VR, gaming, and film.
• Existing 3D face modeling tools are complex and time-consuming, even for experts.
• Current sketch-based tools lack support for amateurs in creating diverse facial shapes and details.

Purpose of the Study:

• To introduce SketchMetaFace, a novel sketching system for amateur users to model high-fidelity 3D faces rapidly.
• To address the challenges of user-friendliness and detail generation in 3D face modeling.
• To develop an efficient and accurate method for mapping 2D sketches to 3D models.

Main Methods:

• Developed SketchMetaFace with a user-friendly interface and curvature-aware strokes for detailed control.
• Introduced Implicit and Depth Guided Mesh Modeling (IDGMM), a learning-based approach fusing mesh, implicit, and depth representations.
• Implemented a coarse-to-fine sketching interface and a data-driven stroke suggestion tool.

Main Results:

• User studies confirmed SketchMetaFace's superiority in ease of use and visual quality compared to existing tools.
• Experimental analysis showed IDGMM achieves a favorable balance between accuracy and computational efficiency.
• The system enables high-fidelity 3D face model generation in minutes.

Conclusions:

• SketchMetaFace significantly lowers the barrier for amateur users to create detailed 3D face models.
• The IDGMM method offers an effective solution for 2D sketch to 3D model conversion.
• This system enhances the accessibility and quality of 3D avatar creation for various applications.