3D Gaussian Head Avatars with Expressive Dynamic Appearances by Compact Tensorial Representations

2025-04-21

AI Summary

Existing methods struggle to simultaneously achieve high dynamic texture fidelity, real-time rendering performance, and storage efficiency in 3D head modeling. This paper proposes a tensorized compact representation framework: neutral geometry and appearance are encoded in static tri-planes, while expressive dynamic details are modeled via lightweight 1D feature curves. To enhance training stability and reconstruction quality, we introduce an adaptive truncation opacity regularization and a class-balanced sampling strategy. Our method integrates 3D Gaussian splatting, 3DMM-based deformation priors, tri-plane encoding, and 1D decoding for high-fidelity, expression-driven dynamic appearance reconstruction. Experiments demonstrate that our approach maintains real-time rendering capability (≥30 FPS) while reducing model size by over 60%. It exhibits strong generalization across diverse input configurations—including single-image and multi-image settings—and enables cross-identity transfer.

Technology Category

• Computer Vision: 3D Computer Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Representation Learning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Recent studies have combined 3D Gaussian and 3D Morphable Models (3DMM) to construct high-quality 3D head avatars. In this line of research, existing methods either fail to capture the dynamic textures or incur significant overhead in terms of runtime speed or storage space. To this end, we propose a novel method that addresses all the aforementioned demands. In specific, we introduce an expressive and compact representation that encodes texture-related attributes of the 3D Gaussians in the tensorial format. We store appearance of neutral expression in static tri-planes, and represents dynamic texture details for different expressions using lightweight 1D feature lines, which are then decoded into opacity offset relative to the neutral face. We further propose adaptive truncated opacity penalty and class-balanced sampling to improve generalization across different expressions. Experiments show this design enables accurate face dynamic details capturing while maintains real-time rendering and significantly reduces storage costs, thus broadening the applicability to more scenarios.

Problem

• Capturing dynamic textures in 3D head avatars efficiently
• Reducing runtime and storage overhead in avatar rendering
• Improving generalization across facial expressions in avatars

Innovation

• Compact tensorial representation for dynamic textures
• Static tri-planes and lightweight 1D feature lines
• Adaptive opacity penalty and balanced sampling