Feature point annotation tool about image-matching-toolbox HOT 10 CLOSED

Thx for your reply, I got it, have a great day!

from image-matching-toolbox.

from image-matching-toolbox.

Actually, I want to test and evalute different image matching methods, so I would labeling some feature point by hand , and use it as groundtruth.

from image-matching-toolbox.

from image-matching-toolbox.

Sorry, I want to evaluate it by use feature points and matchs, but i dont know how to generate them because my data has a lot of light changes

from image-matching-toolbox.

Have you already tried to use the methods supported in this repo and see whether they can give you good matches on your data? Here's an example how you can use them to match any image pair: visualize_matches_on_example_pairs.ipynb.

from image-matching-toolbox.

Not yet, but I will try later, Do you think is make sense for research paper which use some open source image matching methods to get some pre result and select good match result by hand, then use this result as gt for evaluation.

from image-matching-toolbox.

Hm I never seen geometric matching ground-truth hand-picked by human. It is not as accurate as matches computed by geometric transformations. Also is not an efficient way to get GTs since one image pair you can have lots of feature matches.

from image-matching-toolbox.

How to achieve matches computed by geometric transformations？ My data just some image pairs, no depth information and could not do SFM

from image-matching-toolbox.

If you do not have any pose information it's not possible I think. But if your images are from one scene you can run SfM to get poses. In the case if you do not have GTs for those pairs, why you have to stick to those pairs for evaluation? There are common benchmarks to eval matching models and why not use them ? Sorry I don't know details about your final goal. But I guess there's not much more I can help :D.

from image-matching-toolbox.