3D Gaussian Human and Head Avatar Reconstruction
Animatable human/head avatar reconstruction and high-fidelity rendering based on 3D Gaussian representations.
This research direction studies animatable avatar reconstruction with 3D Gaussian representations. The goal is to reconstruct and drive human or head avatars from monocular or few-shot inputs while preserving geometry consistency, appearance quality, and rendering fidelity.
The related work explores deformable Gaussians, topology constraints, barycentric parameterization, layered densification, prior-guided appearance learning, and high-fidelity rendering. Representative projects include 3DGA, CANVAS, and HiFiAvatar.