This page shows complete results for our repeatability evaluation of each feature detector. As described in the paper, this metric computes the percentage of features in image 1 with a close enough detection (in terms of feature support overlap) in image 2 (when warped with the homography). We evaluate repeatability as a function of the number of features detected, where features are ranked either by scale (size), or detection score. The table below (Table 1 in the paper) shows the average repeatability score for the top n = 100 and 200 detections according to scale and score (higher is better).
To see the plots for an image pair, just mouse over a thumbnail; to make the pair "stick", click on the icon.