LEIFR-Net: light estimation for implicit face relight network

Area of Science:

• Computer Vision
• Computer Graphics
• Artificial Intelligence

Background:

• Realistic facial relighting is crucial for digital media.
• Current methods struggle to harmonize facial lighting with environmental conditions.

Purpose of the Study:

• To introduce the Light Estimation for Implicit Face Relight Network (LEIFR-Net).
• To improve the naturalness and environmental alignment of relighted facial images.

Main Methods:

• Estimating global illumination from single images.
• Structurally disentangled facial relighting using pixel-aligned implicit functions.
• Constructing a synthetic dataset using stable diffusion for training.

Main Results:

• LEIFR-Net achieves more harmonious alignment of highlights and shadows with environmental lighting.
• Experimental results demonstrate superior performance compared to contemporary methods.
• The approach enhances the naturalness of relighted facial images across diverse settings.

Conclusions:

• LEIFR-Net represents a novel and effective approach to facial image relighting.
• The method significantly improves the integration of facial lighting with environmental cues.
• This work advances the state-of-the-art in realistic facial relighting.