arXiv:2511.21265v1 Announce Type: new 
Abstract: Learning-based image matching critically depends on large-scale, diverse, and geometrically accurate training data. 3D Gaussian Splatting (3DGS) enables photorealistic novel-view synthesis and thus is attractive for data generation. However, its geometric inaccuracies and biased depth rendering currently prevent robust correspondence labeling. To address this, we introduce MatchGS, the first framework designed to systematically correct and leverage 3DGS for robust, zero-shot image matching. Our approach is twofold: (1) a geometrically-faithful data generation pipeline that refines 3DGS geometry to produce highly precise correspondence labels, enabling the synthesis of a vast and diverse range of viewpoints without compromising rendering fidelity; and (2) a 2D-3D representation alignment strategy that infuses 3DGS' explicit 3D knowledge into the 2D matcher, guiding 2D semi-dense matchers to learn viewpoint-invariant 3D representations. Our generated ground-truth correspondences reduce the epipolar error by up to 40 times compared to existing datasets, enable supervision under extreme viewpoint changes, and provide self-supervisory signals through Gaussian attributes. Consequently, state-of-the-art matchers trained solely on our data achieve significant zero-shot performance gains on public benchmarks, with improvements of up to 17.7%. Our work demonstrates that with proper geometric refinement, 3DGS can serve as a scalable, high-fidelity, and structurally-rich data source, paving the way for a new generation of robust zero-shot image matchers.

تم تقديم إطار جديد يسمى MatchGS لتحسين قدرات 3D Gaussian Splatting (3DGS) في مطابقة الصور بدون تدريب مسبق. يتناول هذا النهج قيود الأساليب السابقة من خلال تحسين هندسة 3DGS ومحاذاة التمثيلات ثنائية وثلاثية الأبعاد، مما يسمح بتوليد تسميات مطابقة دقيقة وتنوع في وجهات النظر دون التضحية بجودة العرض.

Se ha presentado un nuevo marco llamado MatchGS para mejorar las capacidades del 3D Gaussian Splatting (3DGS) en el emparejamiento de imágenes cero-shot. Este enfoque aborda las limitaciones de métodos anteriores al refinar la geometría del 3DGS y alinear representaciones 2D-3D, lo que permite la generación de etiquetas de correspondencia precisas y una variedad de puntos de vista sin sacrificar la calidad de renderizado.

Un nouveau cadre nommé MatchGS a été introduit pour améliorer les capacités du 3D Gaussian Splatting (3DGS) pour le matching d'images zéro-shot. Cette approche s'attaque aux limitations des méthodes précédentes en affinant la géométrie du 3DGS et en alignant les représentations 2D-3D, permettant ainsi la génération d'étiquettes de correspondance précises et de points de vue divers sans compromettre la qualité de rendu.

A new framework named MatchGS has been introduced to enhance the capabilities of 3D Gaussian Splatting (3DGS) for zero-shot image matching. This approach addresses the limitations of previous methods by refining the geometry of 3DGS and aligning 2D-3D representations, enabling the generation of accurate correspondence labels and diverse viewpoints without sacrificing rendering quality.

Unlocking Zero-shot Potential of Semi-dense Image Matching via Gaussian Splatting

Was this article worth reading? Share it

FaceSwapper AI Headshot Generator

Fakeface

Sprello