3D Gaussian Splatting

← Back to Research

Geometry-Aware Style Transfer in 3D Gaussian Splatting

ECCV 2026 (Accepted)  |  Min Hyeok Bang*, Jun Hyeong Kim*, Seung-Wook Kim, and Se-Ho Lee

We present a geometry-aware style transfer framework for 3D Gaussian Splatting (3DGS) that transfers both appearance attributes and geometric structures. Unlike previous 3DGS stylization methods that mainly focus on color-based changes, our method explicitly adapts scene geometry through a decoupled optimization scheme that alternately updates color and geometry parameters.

The core component is geometry-aware contrastive feature matching (GCFM), which combines RGB, depth, and edge cues into a contrastive objective. This multi-modal guidance helps align rendered 3DGS features with a target style image while preserving stable scene-level optimization. Experiments show that the proposed method achieves superior qualitative fidelity and quantitative performance over existing 3DGS-based stylization methods.

Publications

• Min Hyeok Bang, Jun Hyeong Kim, Seung-Wook Kim, and Se-Ho Lee, “Geometry-aware style transfer in 3D Gaussian splatting,” in Proc. European Conference on Computer Vision (ECCV), Sep. 2026 (Accepted). [Paper] [Code] [Project Page]