Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Shanzhi Yin, Bolin Chen, Xinju Wu, Ru-Ling Liao, Jie Chen, Shiqi Wang, Yan Ye

TL;DR

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 is proposed, paving the way for seamless immersive multimedia experiences in meta-verse applications.

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.

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

TL;DR

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 is proposed, paving the way for seamless immersive multimedia experiences in meta-verse applications.

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.

Paper Structure

This paper contains 11 sections, 10 equations, 4 figures.

Table of Contents

  1. Introduction
  2. The Proposed Efficient 3DGS Human Avatar Coding Framework
  3. Experimental Results
  4. Conclusion
  5. References

Figures (4)

  • Figure 1: The detailed structure of proposed human-prior-guided efficient 3D gaussian human avatar compression framework.
  • Figure 2: RD performance comparisons in terms of Rate-PSRN, Rate-SSIM and Rate-LPIPS on ZJU-MoCap and MonoCap datasets
  • Figure 3: Subjective comparisons on ZJU-MoCap nerualbody and MonoCap monocap_source dataset at similar quality
  • Figure 4: Multi-view reconstruction results of proposed method