Paper 2601.03319
CaricatureGS: Exaggerating 3D Gaussian Splatting Faces With Gaussian Curvature
- Published
- Jan 2026
- Research lab
- Independent
- Citations
- 0
- GitHub
- Not linked
01 In brief
Summary
CaricatureGS introduces a method for creating photorealistic, controllable 3D caricature avatars by combining curvature-based geometric deformation with 3D Gaussian Splatting (3DGS).
The pipeline starts with a multiview video, extracts a FLAME mesh, and solves a curvature-weighted Poisson equation to produce an exaggerated mesh.
To train the 3DGS, pseudo-ground-truth caricature images are generated by warping each frame using local affine transformations (LAT) based on mesh correspondences.
A training scheme alternates between real and synthesized supervision, enabling a single Gaussian set to represent both natural and exaggerated avatars.
This approach improves fidelity, supports local edits, and allows continuous control over caricature intensity via an efficient interpolation between original and exaggerated surfaces, with a bounded deviation from exact solutions.
Experiments on the NeRSemble dataset show that CaricatureGS outperforms the SurFhead baseline and a diffusion-based editor in quantitative metrics (CLIP-I, CLIP-D, CLIP-C, DINO, SD) and qualitative results, delivering photorealistic, identity-preserving caricatures with real-time rendering and fine-grained control.
02 From the paper
Abstract
A photorealistic and controllable 3D caricaturization framework for faces is introduced. We start with an intrinsic Gaussian curvature-based surface exaggeration technique, which, when coupled with texture, tends to produce over-smoothed renders. To address this, we resort to 3D Gaussian Splatting (3DGS), which has recently been shown to produce realistic free-viewpoint avatars. Given a multiview sequence, we extract a FLAME mesh, solve a curvature-weighted Poisson equation, and obtain its exaggerated form. However, directly deforming the Gaussians yields poor results, necessitating the synthesis of pseudo-ground-truth caricature images by warping each frame to its exaggerated 2D representation using local affine transformations. We then devise a training scheme that alternates real and synthesized supervision, enabling a single Gaussian collection to represent both natural and exaggerated avatars. This scheme improves fidelity, supports local edits, and allows continuous control over the intensity of the caricature. In order to achieve real-time deformations, an efficient interpolation between the original and exaggerated surfaces is introduced. We further analyze and show that it has a bounded deviation from closed-form solutions. In both quantitative and qualitative evaluations, our results outperform prior work, delivering photorealistic, geometry-controlled caricature avatars.