On the Skinning of Gaussian Avatars

2025-09-14

AI Summary

Linear Blend Skinning (LBS) in Gaussian splatting-based human animation fails to model the nonlinear rotational behavior of Gaussian distributions, leading to rotation artifacts and deformation distortions. Method: We propose Weighted Quaternion Blending (WRB), a novel rotation blending scheme based on weighted quaternion averaging, extending LBS into a nonlinear rotational deformation framework. WRB requires only modifications to skinning weights and quaternion interpolation logic—no renderer reimplementation is needed—ensuring compatibility with arbitrary Gaussian rasterizers. Integrated with forward skinning, Gaussian point-cloud radiance field reconstruction, and pose binding, it forms an end-to-end trainable Gaussian avatar model. Results: Experiments demonstrate that WRB significantly suppresses rotation artifacts, improves animation stability and visual fidelity, achieves high training and inference efficiency, and seamlessly integrates with mainstream rendering engines.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Motion & Tracking
• Humans and AI: Game Design — Virtual Humans, NPCs and Autonomous Characters

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Radiance field-based methods have recently been used to reconstruct human avatars, showing that we can significantly downscale the systems needed for creating animated human avatars. Although this progress has been initiated by neural radiance fields, their slow rendering and backward mapping from the observation space to the canonical space have been the main challenges. With Gaussian splatting overcoming both challenges, a new family of approaches has emerged that are faster to train and render, while also straightforward to implement using forward skinning from the canonical to the observation space. However, the linear blend skinning required for the deformation of the Gaussians does not provide valid results for their non-linear rotation properties. To address such artifacts, recent works use mesh properties to rotate the non-linear Gaussian properties or train models to predict corrective offsets. Instead, we propose a weighted rotation blending approach that leverages quaternion averaging. This leads to simpler vertex-based Gaussians that can be efficiently animated and integrated in any engine by only modifying the linear blend skinning technique, and using any Gaussian rasterizer.

Problem

• Addressing Gaussian deformation artifacts in avatar animation
• Improving rotation handling for non-linear Gaussian properties
• Enabling efficient animation with weighted rotation blending

Innovation

• Weighted rotation blending with quaternion averaging
• Modifying linear blend skinning technique
• Efficient animation with vertex-based Gaussians