Split and Drive: Dual-Axis Disentanglement for Real-Time Gaussian Head Avatars

2026-07-30

AI Summary

Generating animatable and photorealistic head avatars from a single image remains challenging, as existing methods often rely on external tracking and employ unified representations that couple facial regions, thereby limiting expressiveness and rendering quality. This work proposes SpiD, a novel framework featuring a dual-axis decoupling mechanism: the computational axis internalizes per-frame driving signals, eliminating dependence on external tracking, while the feature axis decomposes the head into three geometrically independent Gaussian branches to enable region-specialized modeling. Built upon 3D Gaussian splatting, SpiD achieves end-to-end real-time inference on a single GPU, outperforming state-of-the-art methods in multiple metrics while setting a new standard for speed and visual fidelity.

Abstract

Creating photorealistic animatable head avatars from a single image remains a fundamental challenge in digital human synthesis. While recent 3D Gaussian Splatting methods have achieved promising results, they rely on external tracking pipelines whose latency is excluded from inference measurements. Furthermore, they adopt unified representations that entangle geometrically distinct facial regions, limiting both expressiveness and rendering fidelity. We propose SpiD (Split and Drive), a single-image Gaussian head avatar framework built on two disentanglement axes. The compute axis internalizes per-frame driving, eliminating external tracking dependency at inference. The feature axis decomposes the avatar into three specialized Gaussian branches, each modeling a geometrically distinct facial domain. Extensive experiments demonstrate consistently strong performance against state-of-the-art methods while achieving the fastest inference speed among all compared methods on a single GPU with the complete driving pipeline included.

Problem

• Gaussian Head Avatars
• Single-Image Animation
• Disentanglement
• Real-Time Rendering
• Digital Human Synthesis

Innovation

• Gaussian Splatting
• disentanglement
• real-time avatar
• single-image animation
• facial modeling