Towards Efficient 3D Gaussian Human Avatar Compression: A Prior-Guided Framework

2025-10-12

AI Summary

This paper addresses the challenge of efficiently compressing 3D Gaussian human avatars for video at ultra-low bitrates. Methodologically, it introduces a lightweight, disentangled compression framework that separates appearance from motion: a static canonical portrait—modeled via articulated splatting—serves as the appearance base, while dynamic deformation is represented by only 94 time-varying parameters; these parameters drive differentiable deformation from the canonical space to target views using a human prior template and linear blend skinning (LBS). Appearance is shared across frames, and motion parameters are minimally encoded. The key contribution is the first deep integration of explicit human structural priors into 3D Gaussian compression, substantially reducing redundancy. Experiments demonstrate state-of-the-art rate-distortion performance on mainstream multi-view human portrait datasets, outperforming both conventional 2D/3D codecs and existing learnable 3D Gaussian compression methods.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Learning with Manifolds
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

This paper proposes an efficient 3D avatar coding framework that leverages compact human priors and canonical-to-target transformation to enable high-quality 3D human avatar video compression at ultra-low bit rates. The framework begins by training a canonical Gaussian avatar using articulated splatting in a network-free manner, which serves as the foundation for avatar appearance modeling. Simultaneously, a human-prior template is employed to capture temporal body movements through compact parametric representations. This decomposition of appearance and temporal evolution minimizes redundancy, enabling efficient compression: the canonical avatar is shared across the sequence, requiring compression only once, while the temporal parameters, consisting of just 94 parameters per frame, are transmitted with minimal bit-rate. For each frame, the target human avatar is generated by deforming canonical avatar via Linear Blend Skinning transformation, facilitating temporal coherent video reconstruction and novel view synthesis. Experimental results demonstrate that the proposed method significantly outperforms conventional 2D/3D codecs and existing learnable dynamic 3D Gaussian splatting compression method in terms of rate-distortion performance on mainstream multi-view human video datasets, paving the way for seamless immersive multimedia experiences in meta-verse applications.

Problem

• Compressing 3D human avatar videos at ultra-low bit rates
• Separating appearance and motion for efficient video compression
• Improving rate-distortion performance over conventional 2D/3D codecs

Innovation

• Leveraging compact human priors for compression
• Using canonical-to-target transformation for avatars
• Employing Linear Blend Skinning for temporal coherence