Gaussian Deja-vu: Creating Controllable 3D Gaussian Head-Avatars with Enhanced Generalization and Personalization Abilities

I. Overview

Personalized head avatars often require a lengthy per-person optimization process. Gaussian Deja-vu targets both generalization to a new identity and efficient personalization while preserving explicit control over expression and pose.

The method uses 3D Gaussian rendering to retain real-time performance and high-frequency appearance detail.

II. Key Contributions

• Combines generalizable initialization with efficient identity-specific personalization.
• Builds controllable head avatars around a real-time 3D Gaussian representation.
• Improves personalized avatar quality while reducing the time needed to adapt to a new subject.

III. Methodology

Gaussian Deja-vu learns reusable priors across identities and then adapts the Gaussian avatar representation to a target subject. Facial controls drive the personalized representation while the Gaussian renderer produces novel views in real time.

IV. Main Findings

The WACV evaluation reports faster personalization and improved photorealistic avatar quality, together with controllable expression and pose rendering.

@InProceedings{Yan_2025_WACV,
  author    = {Yan, Peizhi and Ward, Rabab and Tang, Qiang and Du, Shan},
  title     = {Gaussian Deja-vu: Creating Controllable 3D Gaussian Head-Avatars with Enhanced Generalization and Personalization Abilities},
  booktitle = {Proceedings of the Winter Conference on Applications of Computer Vision (WACV)},
  month     = {February},
  year      = {2025},
  pages     = {276--286}
}