Efficient 3D Gaussian Head Avatars for Edge Devices

2026-10-07

AI Summary

This study addresses the prohibitive computational overhead of 3D Gaussian avatar generation, which hinders deployment on edge devices. We propose a lightweight generative architecture that achieves design-level efficiency through parameter-efficient synthesis blocks and depthwise separable convolutions. This approach substantially reduces model complexity without requiring post-hoc compression or quantization, while inherently supporting stylized conditional generation. Experimental results demonstrate that our method reduces FLOPs by 94%, parameters by 70%, and model size by 81%. Notably, it enables real-time inference on mobile devices in GPU-free environments and within web browsers via ONNX Runtime. Overall, this work significantly lowers resource consumption while preserving high-fidelity generation quality.

Abstract

Generative 3D Gaussian head avatars provide high-quality, efficient rendering, but synthesising the Gaussian representation remains computationally expensive, limiting deployment on resource-constrained and edge devices. We introduce an efficient generator architecture for unconditional 3D Gaussian head synthesis, based on a parameter-efficient synthesis block and depth-wise separable convolutions while retaining style-based conditioning. Our architecture reduces generator complexity without requiring model compression or quantisation. Compared with the baseline model, our approach reduces FLOPs by 94%, parameter count by 70%, and model size by 81%, while maintaining competitive generation quality. We further demonstrate practical CPU inference and browser-based execution on mobile devices using ONNX Runtime, enabling 3D Gaussian avatar synthesis without dedicated GPU hardware or application-specific software. In addition to conventional image-quality metrics, we evaluate multi-view consistency, training cost, and deployment performance. Code, trained models, and evaluation tools will be released publicly.

Problem

• 3D Gaussian Head Avatars
• Edge Devices
• Computational Cost
• Resource-constrained Deployment

Innovation

• 3D Gaussian Head Avatars
• Depth-wise Separable Convolutions
• Edge Devices
• Parameter-efficient Synthesis
• ONNX Runtime