Research map / 3D vision / Gaussian splatting
HERA: Hybrid Explicit Representation for Ultra-Realistic Head Avatars
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes HERA, a hybrid explicit head-avatar representation that combines a UV-mapped FLAME mesh (texture and opacity maps) with rigged 3D Gaussian splats, rendered via a stable hybrid depth-sorting alpha-blending pipeline in real time.
Problem
Single-primitive representations struggle with head avatars: 3DGS produces blurry or ribbed artifacts on smooth skin with sharp textures, while meshes cannot capture complex geometry such as hair and eyelashes.
Approach
Fuse UV-mapped mesh and 3DGS splats in one differentiable rasterization by alpha-blending their merged depth-sorted fragments, with a stable sorting rule that projects splat centers onto the mesh depth to avoid splats crossing mesh facets.
Key results
- Novel view synthesis PSNR 33.647 / SSIM 0.910 / LPIPS 0.232 on Multiface, vs 31.728 / 0.882 / 0.308 for GaussianAvatars.
- Novel expression animation PSNR 35.281 / SSIM 0.911 / LPIPS 0.221, vs 34.250 for GaussianAvatars.
- Renders 2048x1334 at >=30 fps on a single NVIDIA RTX A100 GPU.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use HERA's hybrid mesh and 3DGS representation instead of GaussianAvatars | Head avatars · Multiface novel view synthesis | PSNR: 31.728 -> 33.647 | Helped |
| use HERA's hybrid mesh and 3DGS representation instead of GaussianAvatars | Head avatars · Multiface novel expression animation | PSNR: 34.250 -> 35.281 | Helped |
| add the stable depth sorting strategy (ablation) | HERA · Multiface novel view synthesis | PSNR: 32.836 -> 33.647 | Helped |
| add the stable depth sorting strategy (ablation) | HERA · Multiface novel expression animation | PSNR: 34.701 -> 35.281 | Helped |
Limitations
- Relies on FLAME tracking and multi-view capture of each subject
- Stable sorting uses an empirically set threshold (5 cm)
- Evaluated only on Multiface subjects with masked metrics
Experimental setup
Datasets: Multiface
Benchmarks: Multiface
Metrics: PSNR SSIM LPIPS
Baselines: NeRFBlendShape PointAvatar GaussianBlendshapes GaussianAvatars Mesh based Avatars
Builds on: 3D Gaussian Splatting GaussianAvatars FLAME SUGAR
Closest papers by meaning
- MeGA: Hybrid Mesh-Gaussian Head Avatar for High-Fidelity Rendering and Head Editing (CVPR 2025 · similarity 0.91)
- Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians (CVPR 2024 · similarity 0.90)
- Efficient 3D Implicit Head Avatar with Mesh-anchored Hash Table Blendshapes (CVPR 2024 · similarity 0.90)
- GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections (CVPR 2025 · similarity 0.90)
- RGBAvatar: Reduced Gaussian Blendshapes for Online Modeling of Head Avatars (CVPR 2025 · similarity 0.90)
- GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians (CVPR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.