HeadStudio: Text to Animatable Head Avatars
with 3D Gaussian Splatting


Arxiv 2024

Zhenglin Zhou, Fan Ma, Hehe Fan, Yi Yang

ReLER, CCAI, Zhejiang University


Video


Method Overview


Framework of HeadStudio, which integrates FLAME into 3D Gaussian splatting and score distillation sampling. 1) FLAME-based 3D Gaussian Splatting (F-3DGS): each 3D point is rigged to a FLAME mesh, and then rotated, scaled, and translated by the mesh deformation. 2) textbf{FLAME-based Score Distillation Sampling (F-SDS): utilizing FLAME-based fine-grained control signals to guide score distillation. Furthermore, we also introduce additional enhancements, including uniform super-resolution and mesh regularization in F-3DGS, training with animation and denoised score distillation in F-SDS.

Text to Static Avatar Generation

Comparison with the text to static avatar generation methods. HeadStudio excels at producing high-fidelity head avatars, yielding superior results.

Text to Dynamic Avatar Generation

Comparison with the text to dynamic avatar generation methods. HeadStudio provides effective semantic alignment, smooth expression deformation, and real-time rendering.

Easter Egg


Citation

@article{zhou2024headstudio,
  author = {Zhenglin Zhou and Fan Ma and Hehe Fan and Yi Yang},
  title = {HeadStudio: Text to Animatable Head Avatars with 3D Gaussian Splatting},
  journal={arXiv preprint arXiv:2402.06149},
  year={2024}
}