Abstract:Accurately matching local features between a pair of images is a challenging computer vision task. Previous studies typically use attention based graph neural networks (GNNs) with fully-connected graphs over keypoints within/across images for visual and geometric information reasoning. However, in the context of feature matching, considerable keypoints are non-repeatable due to occlusion and failure of the detector, and thus irrelevant for message passing. The connectivity with non-repeatable keypoints not only introduces redundancy, resulting in limited efficiency, but also interferes with the representation aggregation process, leading to limited accuracy. Targeting towards high accuracy and efficiency, we propose MaKeGNN, a sparse attention-based GNN architecture which bypasses non-repeatable keypoints and leverages matchable ones to guide compact and meaningful message passing. More specifically, our Bilateral Context-Aware Sampling Module first dynamically samples two small sets of well-distributed keypoints with high matchability scores from the image pair. Then, our Matchable Keypoint-Assisted Context Aggregation Module regards sampled informative keypoints as message bottlenecks and thus constrains each keypoint only to retrieve favorable contextual information from intra- and inter- matchable keypoints, evading the interference of irrelevant and redundant connectivity with non-repeatable ones. Furthermore, considering the potential noise in initial keypoints and sampled matchable ones, the MKACA module adopts a matchability-guided attentional aggregation operation for purer data-dependent context propagation. By these means, we achieve the state-of-the-art performance on relative camera estimation, fundamental matrix estimation, and visual localization, while significantly reducing computational and memory complexity compared to typical attentional GNNs.

Outlier Robust Feature Correspondence by Learning Based Matching Process

Robust feature matching using guided local outlier factor

Leveraging Local Planar Motion Property for Robust Visual Matching and Localization.

A High Precision Feature Matching Method Based On Geometrical Outlier Removal For Remote Sensing Image Registration

Hierarchical Clustering Matching for Features with Repetitive Patterns in Visual Odometry

CSR-Net++: Rethinking Context Structure Representation Learning for Feature Matching

Extended Neighborhood Consensus With Affine Correspondence for Outlier Filtering in Feature Matching

Learning Bipartite Graph Matching for Robust Visual Localization.

A Hypergraph Matching Framework for Refining Multi-source Feature Correspondences.

A More Robust Feature Correspondence for more Accurate Image Recognition

Feature Matching via Topology-Aware Graph Interaction Model

Learning Relaxed Neighborhood Consistency for Feature Matching

A Novel Image Registration Method Based on Geometrical Outlier Removal

Learning to find reliable correspondences with local neighborhood consensus

Learning Feature Matching via Matchable Keypoint-Assisted Graph Neural Network

A Robust Algorithm For Feature Point Matching

An Accurate and Robust Registration Framework Based on Outlier Removal and Feature Point Adjustment for Remote Sensing Images

Robust feature point matching based on geometric consistency and affine invariant spatial constraint

Robust Outlier Rejection for 3D Registration with Variational Bayes

iMatching: Imperative Correspondence Learning

Generalized Correspondence Matching via Flexible Hierarchical Refinement and Patch Descriptor Distillation