In this work, we address the problem of refining the geometry of local image features from
multiple views without known scene or camera geometry. Current approaches to local
feature detection are inherently limited in their keypoint localization accuracy because they
only operate on a single view. This limitation has a negative impact on downstream tasks
such as Structure-from-Motion, where inaccurate keypoints lead to large errors in
triangulation and camera localization. Our proposed method naturally complements the …

This supplementary material provides the following information: Section 1 explains how we
determined the match filtering thresholds for the learned methods. Section 2 contains the
additional results mentioned in the main paper (eg, ETH3D [11] triangulation results on each
individual dataset and independent results of each method on the Local Feature Evaluation
Benchmark [10]) as well as some qualitative examples before and after refinement. Section
3 presents an ablation study for both the two-view and the multi-view refinement procedure …