Multi-StyleGS: Stylized Gaussian Splatting with Multiple Styles

2025-04-11

AAAI Conference on Artificial Intelligence

AI Summary

This work addresses the challenge of multi-style artistic editing for 3D Gaussian Splatting scenes—requiring cross-view consistency, memory efficiency, and support for both automatic and manual local style transfer. Methodologically, it introduces a novel bipartite graph matching–driven mechanism to establish correspondences between local styles and scene regions; incorporates semantic style loss and multi-scale local-global feature matching to achieve object-level style disentanglement and enhanced view consistency; and employs memory-aware optimization to significantly improve texture detail and color fidelity under constrained GPU memory. Experiments demonstrate superior performance over state-of-the-art methods on multi-style editing tasks, enabling flexible and robust interactive 3D artistic creation. Key contributions include the first formulation of style-region correspondence via bipartite matching in Gaussian splatting, effective decoupling of stylistic and geometric attributes while preserving view coherence, and efficient rendering with high visual fidelity at low memory cost.

Technology Category

• Computer Vision: 3D Computer Vision
• Humans and AI: Game Design — Procedural Content Generation & Storytelling
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

In recent years, there has been a growing demand to stylize a given 3D scene to align with the artistic style of reference images for creative purposes. While 3D Gaussian Splatting (GS) has emerged as a promising and efficient method for realistic 3D scene modeling, there remains a challenge in adapting it to stylize 3D GS to match with multiple styles through automatic local style transfer or manual designation, while maintaining memory efficiency for stylization training. In this paper, we introduce a novel 3D GS stylization solution termed Multi-StyleGS to tackle these challenges. In particular, we employ a bipartite matching mechanism to automatically identify correspondences between the style images and the local regions of the rendered images. To facilitate local style transfer, we introduce a novel semantic style loss function that employs a segmentation network to apply distinct styles to various objects of the scene and propose a local-global feature matching to enhance the multi-view consistency. Furthermore, this technique can achieve memory-efficient training, more texture details and better color match. To better assign a robust semantic label to each Gaussian, we propose several techniques to regularize the segmentation network. As demonstrated by our comprehensive experiments, our approach outperforms existing ones in producing plausible stylization results and offering flexible editing.

Problem

• Stylizing 3D Gaussian Splatting with multiple artistic styles
• Achieving memory-efficient training for multi-style 3D scene modeling
• Enhancing multi-view consistency and local style transfer accuracy

Innovation

• Bipartite matching for style-region correspondences
• Semantic style loss with segmentation network
• Local-global feature matching for consistency