Pose graph optimization improves camera positioning from unordered photos
Learning Scene Coordinate Reconstruction from Unposed Images via Pose Graph Optimization
Computer Vision and Pattern Recognition
Summary
Estimating where cameras were when photos were taken is tricky, especially if photos are in random order and the scene is complicated. The authors show that combining traditional geometry-based approaches with AI predictions leads to better results. Their method uses connections between images to refine camera positions and reduces errors caused by confusing scenes. This improves the overall accuracy and reliability of reconstructing 3D scenes from pictures.
What this means in practice
- •For augmented reality developers: Enhance AR experiences by providing more accurate and robust camera positioning from unordered images without prior calibration data.
- •For drone mapping teams: Improve mapping accuracy and reduce pose errors when reconstructing scenes from aerial drone imagery collected in unstructured flight patterns.
Authors
Tze Ho Elden Tse, Jizong Peng, Angela Yao
Abstract
Learning-based structure-from-motion methods such as ACE-Zero have demonstrated strong performance in estimating cameraposes and scene coordinates from unordered image collections without requiring ground truth supervision. However, the lack of global and multi-view consistency constraints in ACE-Zero can lead to pose drift and misalignment, particularly in complex or ambiguous scenes. In this work, we propose a hybrid framework that integrates pose graph optimization (PGO) into ACE-Zero to refine camera poses and suppress incorrect refinements. We construct pose graphs directly from ACE-Zero outputs by extracting relative pose constraints from predicted scene coordinates. Furthermore, we introduce an uncertainty-aware optimization strategy by estimating confidence scores using geometric priors, including epipolar and optical flow consistencies across views. Our approach improves the robustness and accuracy of pose estimation, demonstrating that global geometric reasoning can effectively complement learning-based inference in structure-from-motion.