DynAvatar: Dynamic 3D Head Avatar Deformation With Expression Guided Gaussian Splatting.

Area of Science:

• Computer Graphics
• Computer Vision
• Human-Computer Interaction

Background:

• Generating high-fidelity 3D head avatars is crucial for immersive technologies like virtual reality and gaming.
• Existing methods struggle with precise non-rigid facial deformation and semantically controlled expression synthesis.

Purpose of the Study:

• To introduce DynAvatar, a novel framework for creating photorealistic and emotionally resonant 3D head avatars.
• To enable fine-grained, anatomically meaningful facial animation with semantic control.

Main Methods:

• DynAvatar integrates an expression-guided Gaussian deformation module with the 3D Gaussian splatting pipeline.
• A spatial context embedding mechanism preserves semantic coherence and spatial consistency during expression generation.

Main Results:

• DynAvatar significantly improves visual realism and expression fidelity compared to state-of-the-art methods.
• The framework demonstrates superior rendering quality across diverse datasets.

Conclusions:

• DynAvatar offers a robust solution for generating high-quality, expressive 3D head avatars.
• The method advances the state-of-the-art in 3D avatar creation for immersive applications.