vehicle feature

[Nestart]

hello,can you give me some suggestions on why this method is so insensitive to vehicles?

[Nestart]

[ducha-aiki]

What do you trying to achieve? I think that makes total sense that static scene geometry is easier to find and the method outputs the model with the largest number of correspondences

[Nestart]

What do you trying to achieve? I think that makes total sense that static scene geometry is easier to find and the method outputs the model with the largest number of correspondences

Thanks for your quick reply.In fact,i want to find the correspondences between the car in image1 and image2,and when i use sift in opencv-python to match ,it finds many correspondences on the car .

[ducha-aiki]

Then one way would be to crop the car part :)
When you use open-cv sift, what you find is just correspondences. When you use MODS, you find the correspondences, which are consistent with some geometrical model - homography or epipolar geometry. It is by design.
In fact, you can mimic MODS result with opencv sift, if you add homography estimation part:

https://cw.fel.cvut.cz/b192/courses/mpv/labs/2_correspondence_problem/start

[Nestart]

Then one way would be to crop the car part :)
When you use open-cv sift, what you find is just correspondences. When you use MODS, you find the correspondences, which are consistent with some geometrical model - homography or epipolar geometry. It is by design.
In fact, you can mimic MODS result with opencv sift, if you add homography estimation part:

https://cw.fel.cvut.cz/b192/courses/mpv/labs/2_correspondence_problem/start

It helps me a lot,i will do more try,Thanks!