GALA: Guided Attention with Language Alignment for Open Vocabulary Gaussian Splatting

Elena Alegret1,2,6,*, Kunyi Li1,4,*, Sen Wang1,4, Siyun Liang5, Michael Niemeyer3, Stefano Gasperini1,4,7, Nassir Navab1,4 Federico Tombari1,3
* Equal Contribution, 1Technical University of Munich, 2Universitat Politecnica de Catalunya, 3Google, 4Munich Center for Machine Learning, 5University of Tubingen, 6ETH Zurich, 7Visualais

3DV 2025

Paper arXiv Code

Abstract

We introduce GALA, a novel framework for open-vocabulary 3D scene understanding with 3D Gaussian Splatting (3DGS). GALA distills a scenespecific 3D instance feature field via self-supervised contrastive learning. To extend to generalized language feature fields, we introduce the core contribution of GALA, a crossattention module with two learnable codebooks that encode view-independent semantic embeddings. This design not only ensures intra-instance feature similarity but also supports seamless 2D and 3D open-vocabulary queries. It reduces memory consumption by avoiding per-Gaussian high-dimensional feature learning. Extensive experiments on real-world datasets demonstrate GALA’s remarkable open-vocabulary performance on both 2D and 3D.

GALA

GALA teaser

2D Open-Vocabulary Query

GALA teaser

3D Open-Vocabulary Segmentation

GALA teaser

BibTeX

@InProceedings{alegret2025gala, 
      author = {Elena Alegret* and Kunyi Li* and Sen Wang and Siyun Liang and Michael Niemeyer and Stefano Gasperini and Nassir Navab and Federico Tombari}, 
      title = {GALA: Guided Attention with Language Alignment for Open Vocabulary Gaussian Splatting}, 
      booktitle = {International Conference on 3D Vision (3DV)}, 
      year = {2026},
    }