Semantic inpainting improves animatable head avatars from single images
Learning Semantic Inpainting for Animatable Gaussian Head Avatars
Computer Vision and Pattern Recognition
Summary
Making realistic digital head avatars from just one photo is hard because parts of the face are hidden. The authors came up with a method that fills in those missing parts by using the known patterns and symmetry of faces. Their technique represents a face using special 3D points called Gaussians, stacking several to show fine details. This lets the avatars look complete and keeps the person’s identity clear, while allowing the avatar to be animated naturally from different angles.
What this means in practice
- •For game developers: Generate animatable 3D head avatars from single photos for games without needing multiple images or scans.$Commercial implications: Enables game studios to create personalized character avatars at scale from just one image, improving user experience and reducing production costs.
- •For virtual reality content creators: Create realistic head avatars for VR experiences that animate smoothly from limited input images.
Authors
Pilseo Park, Fizza Rubab, Yiying Tong
Abstract
We present SInGA, a novel method for learning Semantic Inpainting for animatable Gaussian head Avatars from a single image. Existing avatar approaches often rely on multi-view observations and lack effective handling of unobserved regions in single-view settings, limiting their applicability in such scenarios. To address this, we propose a semantic inpainting framework defined in UV space for completing unobserved facial regions. Our key insight lies in the structured topology of the UV representation, which provides consistent spatial correspondences and enables reliable completion of identity-specific features using the inherent symmetry cues of human faces. We extract features from observed regions and use them to complete unobserved regions. The completed representation is then used to regress Gaussian attributes, effectively performing Gaussian inpainting. In addition, instead of relying on a single Gaussian at each surface or pixel location, we stack multiple Gaussians to enhance detail. The resulting avatar generalizes across identities without requiring per-identity optimization and can be animated with driving inputs. Experimental results show that our method generates high-quality head avatars with improved completeness and identity preservation, while supporting realistic animation and consistent rendering from unobserved views.