search

b51 / FeatureTracker

Features detection, match and track module. Supported features: orb and superpoints
MIT License
30 stars 6 forks source link

Can u explain ur matching strategy in superpointmatcher? #1

Open SYSUGrain opened 4 years ago

SYSUGrain commented 4 years ago

Hi, I see ur code recently, and it's an effective c++ version code about superpoint. But I can't understand the matching method in ur code. Why dot product can represent similarity between descriptors?

Alex-Beh commented 4 months ago

My understanding:

ORB is a binary descriptor so we can use hamming distance to measure the similarity. However the descriptor of Superpoint/SIFT is floating-point descriptor, so we will use euclidean distance.

The idea behind the dot product should be Euclidean Distance Matrices.

You can refer to the explanation here:

1.Euclidean Distance Matrix 2.Euclidean Distance Matrices: A Short Walk Through Theory, Algorithms and Applications

@b51 Please correct me if I was wrong.