Get a score like result

[simonhard]

Hey, thanks for all your great work regarding VPR/Image matching etc.

I want to compare two images and get a result, how much image 1 matches 2. Any kind of metric.

I thought about something like

total_possible_keypoints = min(len(result['kpts0']), len(result['kpts1']))
matching_score = (num_inliers / total_possible_keypoints) * 100

Or ensuring the homography matrix is close to the identity matrix or the number of inliers surpasses a certain threshold...

Any hints are helpful. Thanks :)

[gmberton]

Usually we rely on the number of inliers (num_inliers) to estimate if two images match or not, which is also what we did in the paper. The only thing to take into account is that the number of inliers is dependent on the method and not absolute (e.g. RoMa produces a lot of inliers while SIFT-LG produces few)