F3G-Avatar : Face Focused Full-body Gaussian Avatar

Willem Menu; Erkut Akdag; Pedro Quesado; Yasaman Kashefbahrami; Egor Bondarev

cs.CVApr 10, 2026

F3G-Avatar : Face Focused Full-body Gaussian Avatar

Authors: Willem Menu, Erkut Akdag, Pedro Quesado, Yasaman Kashefbahrami, Egor Bondarev

Organizations: AIMS Group, Department of Electrical Engineering, Eindhoven University of Technology

Abstract

Existing full-body Gaussian avatar methods primarily optimize global reconstruction quality and often fail to preserve fine-grained facial geometry and expression details. This challenge arises from limited facial representational capacity that causes difficulties in modeling high-frequency pose-dependent deformations. To address this, we propose F3G-Avatar, a full-body, face-aware avatar synthesis method that reconstructs animatable human representations from multi-view RGB video and regressed pose/shape parameters. Starting from a clothed Momentum Human Rig (MHR) template, front/back positional maps are rendered and decoded into 3D Gaussians through a two-branch architecture: a body branch that captures pose-dependent non-rigid deformations and a face-focused deformation branch that refines head geometry and appearance. The predicted Gaussians are fused, posed with linear blend skinning (LBS), and rendered with differentiable Gaussian splatting. Training combines reconstruction and perceptual objectives with a face-specific adversarial loss to enhance realism in close-up views. Experiments demonstrate strong rendering quality, with face-view performance reaching PSNR/SSIM/LPIPS of 26.243/0.964/0.084 on the AvatarReX dataset. Ablations further highlight contributions of the MHR template and the face-focused deformation. F3G-Avatar provides a practical, high-quality pipeline for realistic, animatable full-body avatar synthesis.

Figures & tables

Figure 2: Overview of F3G-Avatar. (a) MHR clothed body template. (b) Global Canonical Deformation (Body): front/back body positional maps are processed by the BodyUNet to predict pose-dependent body Gaussians. (c) Face-focused Deformation: head positional maps drive three StyleUNets to predict positional, color, and auxiliary face attributes. (d) Region-aware Reconstruction : The two branches are fused, posed via LBS, rendered with 3DGS, and optimized with reconstruction losses and a face-specific adversarial loss.

Figure 3: Visualization of the canonical face model construction from the MHR template and head positional maps.

	Novel View (Body)			Novel View (Head)
Model	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
AvatarRex [ 40 ]	23.248	0.957	0.065	-	-	-
3DGS-Avatar [ 28 ]	28.784	0.951	0.042	24.972	0.943	0.121
GaussianAvatar [ 6 ]	26.950	0.939	0.041	24.011	0.933	0.144
AnimatableGaussians [ 16 ]	30.361	0.968	0.034	25.671	0.957	0.114
Ours (F3G-Avatar)	30.214	0.970	0.032	26.243	0.964	0.084

Table 1: Quantitative comparison of full-body and face-focused novel-view synthesis on the AvatarReX [ 40 ] dataset.

	Novel View (Body)			Novel View (Head)
Model	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
TAVA [ 13 ]	26.61	0.968	0.032	-	-	-
Ani-NeRF [ 25 ]	22.53	0.964	0.034	20.14	0.931	0.102
AnimatableGaussians [ 16 ]	30.614	0.980	0.029	26.434	0.953	0.071
Ours (F3G-Avatar)	30.311	0.981	0.026	26.934	0.961	0.062

Table 2: Quantitative comparison of full-body and face-focused novel-view synthesis on the THuman4.0 [ 39 ] dataset.

Figure 4: F3G-Avatar displays state-of-the-art rendering quality by delivering improved facial details.

Metric	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
w/o Face-focused Deformation	25.721	0.951	0.107
w/o MHR template	26.015	0.963	0.090
w/o $\mathcal{L_{\text{adv}}}$	27.010	0.959	0.091
Full model	26.243	0.964	0.084

Table 3: Ablation study on AvatarReX head novel-view metrics.

Metric	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Time (ms)
Input Resolution
(128 x 128)	24.554	0.939	0.101	24.34 $\pm$ 1.12
(256 x 256)	26.774	0.956	0.086	61.41 $\pm$ 0.87
Model Variations
$(n_{\text{mlp}},\mathrm{cm})$ : (2, 1)	25.340	0.941	0.087	47.43 $\pm$ 1.11
$(n_{\text{mlp}},\mathrm{cm})$ : (4, 1)	26.243	0.964	0.084	56.47 $\pm$ 1.16

Table 4: Ablation study on model variants with runtime Face-focused Deformation (Time).

Explore similar work

FlexiAvatar: Unified 3D Gaussian Human Avatars Under Arbitrary Body Visibility
Jul 21, 2026Yihalem Yimolal Tiruneh, Muhammad Salman Ali, Uyoung Jeong +53D Gaussian Splatting 3D Reconstruction
UNIST, South Korea · University of Aberdeen, UK · University College London, UK +1
AGORA: Adversarial Generation Of Real-time Animatable 3D Gaussian Head Avatars
Dec 6, 2025Ramazan Fazylov, Sergey Zagoruyko, Aleksandr Parkin +23D Gaussian Splatting 3D Asset Generation
Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE · Polynome AI, Dubai, UAE · MTS AI, Moscow, Russia
FFAvatar: Few-Shot, Feed-Forward, and Generalizable Avatar Reconstruction
May 14, 2026Thuan Hoang Nguyen, Jiahao Luo, Yinyu Nie +33D Gaussian Splatting 3D Reconstruction
Snap Inc. · MBZUAI · University of California, Santa Cruz

Jul 21, 2026J/K move · Enter open · S save

FlexiAvatar: Unified 3D Gaussian Human Avatars Under Arbitrary Body Visibility

Yihalem Yimolal Tiruneh, Muhammad Salman Ali, Uyoung Jeong, Muneeb A. Khan+4

UNIST, South Korea · University of Aberdeen, UK · University College London, UK+1

3D Gaussian Splatting 3D Reconstruction Monocular 3D Reconstruction 3D Avatar Generation 3D Human Reconstruction

Open paper AskPDF arXiv

Reconstructing animatable 3D human avatars from monocular video is a fundamental problem in computer vision with broad applications in AR/VR and digital content creation. Existing approaches typically couple parametric body models with neural rendering or 3D Gaussian splatting and optimize all body regions jointly from short videos, which often degrades fidelity in the visible areas. To overcome this limitation, we introduce FlexiAvatar, a unified framework that explicitly optimizes only the visible body regions, effectively eliminating artifacts arising from unobserved limbs. Our method integrates occlusion-robust SMPL-X tracking with part-specific residual refinement to capture high-frequency geometric and appearance details. To complete entirely unseen regions (e.g., back views), we leverage a diffusion-based approach to generate texture consistent with the observed appearance. Experiments on full-body (NeuMan, ZJU-MoCap, WildAvatar), upper/half-body (talk-show clips), and head-only (INSTA) inputs show that FlexiAvatar delivers consistently higher reconstruction quality, outperforming state-of-the-art methods by an average PSNR improvement of approximately 3% across datasets. Finally, by restricting optimization to observed regions, our method reduces the effective number of Gaussians that must be optimized and rendered, leading to reduced runtime and memory overhead in partial-visibility scenarios.

Abstract

Explore similar work

FlexiAvatar: Unified 3D Gaussian Human Avatars Under Arbitrary Body Visibility

AGORA: Adversarial Generation Of Real-time Animatable 3D Gaussian Head Avatars

FFAvatar: Few-Shot, Feed-Forward, and Generalizable Avatar Reconstruction