Research map / 3D vision / Face and shape
GPHM: Gaussian Parametric Head Model for Monocular Head Avatar Reconstruction
Yuelang Xu, Zhaoqi Su, Qiang Wu, Yebin Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| initialize Gaussians from the guiding geometry model instead of FLAME (ablation) | GPHM · Initialization ablation | PSNR: 25.7 -> 28.0 | Improved |
| initialize Gaussians from the guiding geometry model instead of FLAME (ablation) | GPHM · Initialization ablation | LPIPS: 0.109 -> 0.085 | Improved |
| use 3D Gaussians instead of mesh (ablation) | GPHM · Representation ablation | PSNR: 15.7 -> 27.0 | Improved |
| add super-resolution network to mesh (ablation) | GPHM · Representation ablation | PSNR: 15.7 -> 17.3 | Improved |
| add super-resolution network to Gaussians (ablation) | GPHM · Representation ablation | PSNR: 27.0 -> 29.3 | Improved |
| fine-tune motion-related networks in addition to Gaussians (ablation) | GPHMv2 · Self reenactment | PSNR: 28.1 -> 28.9 | Improved |
| add dynamic generator f_dyn (ablation) | GPHMv2 · Self reenactment | LPIPS: 0.044 -> 0.041 | Improved |
| add dynamic generator f_dyn (ablation) | GPHMv2 · Self reenactment | SSIM: 0.93 -> 0.94 | No clear effect |
| use GPHMv2 instead of FlashAvatar | GPHMv2 · Self reenactment | PSNR: 28.8 -> 28.9 | Improved |
| use GPHMv2 instead of FlashAvatar | GPHMv2 · Self reenactment | LPIPS: 0.051 -> 0.041 | Improved |
| use GPHMv2 fitting instead of PanoHead | GPHMv2 · Single image fitting | PSNR: 29.1 -> 30.3 | Improved |
| add LivePortrait-synthesized expression condition images (ablation) | GPHMv2 · Cross-identity reenactment | appearance leakage (qualitative) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 5 of 5 on most shared results.
- PanoHead — PanoHead: Geometry-Aware 3D Full-Head Synthesis in 360deg (CVPR 2023) · better on 2 of 3 shared results
- HeadNeRF — HeadNeRF: A Real-Time NeRF-Based Parametric Head Model (CVPR 2022) · better on 3 of 3 shared results
- MoFaNeRF — MoFaNeRF: Morphable Facial Neural Radiance Field (ECCV 2022) · better on 3 of 3 shared results
- FlashAvatar — FlashAvatar: High-fidelity Head Avatar with Efficient Gaussian Embedding (CVPR 2024) · better on 2 of 2 shared results
- SplattingAvatar — SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting (CVPR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians (CVPR 2024 · similarity 0.92)
- Generalizable and Animatable Gaussian Head Avatar (NeurIPS 2024 · similarity 0.92)
- Gaussian Eigen Models for Human Heads (CVPR 2025 · similarity 0.92)
- GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections (CVPR 2025 · similarity 0.91)
- GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians (CVPR 2024 · similarity 0.91)
- 3D Gaussian Head Avatars with Expressive Dynamic Appearances by Compact Tensorial Representations (CVPR 2025 · similarity 0.91)
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.
