BrushGaussian: Brushstroke-Based Stylization for 3D Gaussian Splatting
Zhenhua Xiang, Chun Xie, Itaru Kitahara · Applied Sciences · 2025
We present a method for enhancing 3D Gaussian Splatting primitives with brushstroke-aware stylization. Previous approaches to 3D style transfer are typically limited to color or texture modifications, lacking an understanding of artistic shape deformation. In contrast, we focus on individual 3D Gaussian primitives, exploring their potential to enable style transfer that incorporates both color- and brushstroke-inspired local geometric stylization. Specifically, we introduce additional texture features for each Gaussian primitive and apply a texture mapping technique to achieve brushstroke-like geometric effects in a rendered scene. Furthermore, we propose an unsupervised clustering algorithm to efficiently prune redundant Gaussians, ensuring that our method seamlessly integrates with existing 3D Gaussian Splatting pipelines. Extensive evaluations demonstrate that our approach outperforms existing baselines by producing brushstroke-aware artistic renderings with richer geometric expressiveness and enhanced visual appeal.