Neural 3D Face Shape Stylization Based on Single Style Template via Weakly Supervised Learning

Area of Science:

• Computer Graphics
• Artificial Intelligence

Background:

• 3D face shape stylization transforms realistic faces into stylized versions (e.g., cartoons).
• Traditional deformation transfer methods are slow and require re-optimization for each new face.

Purpose of the Study:

• To develop a faster and more efficient method for 3D face shape stylization.
• To address the limitations of traditional deformation transfer techniques.

Main Methods:

• Proposed a neural network-based approach for 3D face shape stylization.
• Employed weakly supervised learning and a novel template-guided mesh smoothing regularization.
• Modeled the task as a deformation transfer problem.

Main Results:

• Achieved stylization quality comparable to traditional methods (average Chamfer Distance ~0.01 mm).
• Significantly improved processing speed, approximately 3,000 times faster than traditional methods.
• Eliminated the need for paired training data.

Conclusions:

• The proposed neural network method offers an efficient and practical solution for 3D face shape stylization.
• This learning-based deformation transfer method is the first of its kind for 3D face stylization.
• The method preserves template structure while enabling rapid stylization.