Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Relational Epipolar Graphs (REGs) for robust relative camera pose estimation, representing matched keypoints as nodes in a graph and leveraging message passing to enforce geometric consistency. By training a graph network to estimate rotation, translation, and the Essential Matrix while minimizing a composite loss function, the method achieves improved robustness against noisy correspondences and large baseline variations. Experiments on indoor and outdoor datasets demonstrate superior performance compared to both classical and learning-guided relative pose estimation techniques.
Forget RANSAC – relational reasoning over epipolar graphs offers a surprisingly robust new approach to camera pose estimation, even with dense noise and large baselines.
A key component of Visual Simultaneous Localization and Mapping (VSLAM) is estimating relative camera poses using matched keypoints. Accurate estimation is challenged by noisy correspondences. Classical methods rely on stochastic hypothesis sampling and iterative estimation, while learning-based methods often lack explicit geometric structure. In this work, we reformulate relative pose estimation as a relational inference problem over epipolar correspondence graphs, where matched keypoints are nodes and nearby ones are connected by edges. Graph operations such as pruning, message passing, and pooling estimate a quaternion rotation, translation vector, and the Essential Matrix (EM). Minimizing a loss comprising (i) $\mathcal{L}_2$ differences with ground truth (GT), (ii) Frobenius norm between estimated and GT EMs, (iii) singular value differences, (iv) heading angle differences, and (v) scale differences, yields the relative pose between image pairs. The dense detector-free method LoFTR is used for matching. Experiments on indoor and outdoor benchmarks show improved robustness to dense noise and large baseline variation compared to classical and learning-guided approaches, highlighting the effectiveness of global relational consensus.