ImFace++: A Sophisticated Nonlinear 3D Morphable Face Model With Implicit Neural Representations.

Area of Science:

• Computer Vision
• Computer Graphics
• Machine Learning

Background:

• Accurate 3D face representations are crucial for computer vision and graphics.
• Current models struggle with data discretization and linearity, limiting identity and expression capture.

Purpose of the Study:

• To present ImFace++, a novel 3D morphable face model leveraging implicit neural representations.
• To overcome limitations in current 3D face modeling for improved identity and expression accuracy.

Main Methods:

• ImFace++ utilizes disentangled deformation fields for identity and expression.
• Incorporates a refinement displacement field for fine-grained facial details.
• Employs a Neural Blend-Field for enhanced representation via adaptive local field blending.

Main Results:

• ImFace++ demonstrates significant advancements in 3D face reconstruction fidelity.
• Achieves state-of-the-art accuracy in point-to-point correspondence across diverse facial shapes.
• Improved learning strategy extends expression embeddings for broader variations.

Conclusions:

• ImFace++ offers a sophisticated and continuous 3D morphable face model.
• The model significantly improves the state-of-the-art in 3D face analysis.
• Enables more precise capture of identity and expression clues in facial data.