LC-NeRF: Local Controllable Face Generation in Neural Radiance Field

Area of Science:

• Computer Vision
• Computer Graphics
• Artificial Intelligence

Background:

• Neural Radiance Fields (NeRF) have advanced 3D face generation, achieving high visual quality and consistency.
• Existing NeRF-based methods often lack fine-grained control over specific facial regions, limiting independent editing of features like the mouth, nose, or hair.

Purpose of the Study:

• To introduce LC-NeRF, a novel framework designed to enhance local controllability in NeRF-based 3D face generation.
• To enable precise geometry and texture control of individual facial regions.

Main Methods:

• LC-NeRF utilizes a Local Region Generators Module (LRGM) to model distinct facial areas as independent NeRFs.
• A Spatial-Aware Fusion Module (SAFM) integrates these independent NeRFs into a cohesive 3D face representation.
• Modifications to latent codes associated with each generator allow for targeted control over specific facial regions.

Main Results:

• Evaluations demonstrate superior local controllability of LC-NeRF compared to state-of-the-art 3D-aware face generation methods.
• Perception studies confirm LC-NeRF's outperformance in image quality, face consistency, and editing effectiveness.
• The method shows strong performance in downstream applications like real image and text-driven facial editing.

Conclusions:

• LC-NeRF significantly improves local controllability in 3D face generation using NeRF.
• The proposed architecture offers precise control over facial geometry and texture, outperforming existing methods in quality and editing capabilities.