GS-CPE: Unified 6-Degree-of-Freedom Camera Pose Estimation via 3D Gaussian Splatting

2026-08-11

AI Summary

Existing visual localization methods struggle to simultaneously achieve high accuracy and strong generalization in 6-degree-of-freedom camera pose estimation. This work proposes GS-CPE, the first end-to-end differentiable coarse-to-fine pose estimation framework based on 3D Gaussian splatting. It begins with a retrieval-guided geometric approach to obtain an initial pose estimate, followed by a refinement stage that leverages a multi-scale visibility-aware RGB image warping loss and adaptive re-rendering for precise optimization. By unifying geometric constraints with learning-driven strategies, GS-CPE achieves state-of-the-art performance across multiple benchmarks—including 7Scenes, Cambridge Landmarks, FAST-LIVO2, and a custom dataset—demonstrating significantly improved pose accuracy and cross-scene generalization capability.

Abstract

Despite substantial progress in visual localization, from scene coordinate regression to direct camera pose regression, achieving both robust generalization and high accuracy remain challenging. This study introduces GS-CPE (Gaussian Splatting based Camera Pose Estimation), a coarse-to-fine framework for 6-DoF camera pose estimation that unifies geometry-based coarse pose estimation with robust 3D Gaussian Splatting (3DGS) warping based pose refinement. GS-CPE first estimates a coarse pose via retrieval-guided geometric pose estimation on a 3DGS scene representation, then refines it by minimizing a visibility aware masked RGB warping objective in a multi-scale optimization framework, with adaptive re-rendering. Extensive experiments on indoor and outdoor benchmarks including 7Scenes, Cambridge Landmarks, FAST-LIVO2 datasets, and a custom dataset demonstrate state-of-the-art performance, consistently outperforming in both accuracy and generalization.

Problem

• camera pose estimation
• 6-DoF
• visual localization
• generalization
• accuracy

Innovation

• 3D Gaussian Splatting
• 6-DoF camera pose estimation
• coarse-to-fine optimization
• visibility-aware warping
• scene representation