imagematching does not work with akaze features

[massimo-tech3d]

I've searched the issues and I think I'm the only one having this problem, so I guess if there's any misconfiguration on my PC:
ImageMatching always fails, regardless of the dataset, when AKAZE features are enabled, either AKAZE alone or in combination with SIFT.

The error reported is "Can't find descriptor of view 18975381 in: " but of course the directory contains both 18975381.akaze.desc and 18975381.akaze.feat files, and they are not empty.

any suggestion how to fix this ?
thanks

[fabiencastan]

Hi,
The vocabularytree used has been learned on SIFT features, so currently the ImageMatching is hardcoded to use SIFT descriptors only.
So if you use AKAZE alone you cannot use the ImageMatching. If your dataset is not too large, you can disable the usage of the vocabularytree by setting Nb Matches to 0 (but then it will compute the matching between all images pairs.

But you should be able to use the ImageMatching if SIFT and AKAZE are used in combination (both checked on the FeatureExtraction node). If it doesn't work, please provide the full log in this particular case.

[massimo-tech3d]

understood, thanks. And you're right, it works with AKAZE if SIFT is also enabled, just tried. Must have had a brain memory failure :-)

[fabiencastan]

Great!
If you want to create a new voctree dedicated for akaze, the code to create a new voctree is available in the repository. Just need to download lots of images and have computation ressources. Then you need to try and evaluate multiple sizes for the voctree to find a good trade-off between size and quality. We can setup a confcall if you (or someone else here) need to go that way.
Best,

[massimo-tech3d]

can you give an idea about how many images is lots of images ? I would give it a try in the next days

[natowi]

can you give an idea about how many images is lots of images ?

This are voctrees for colmap

  Vocabulary tree with 32K visual words
   (for small-scale image collections, i.e. 100s to 1,000s of images)
   Vocabulary tree with 256K visual words
   (for medium-scale image collections, i.e. 1,000s to 10,000s of images)
   Vocabulary tree with 1M visual words
   (for large-scale image collections, i.e. 10,000s - 100,000s of images)

The original publication uses datasets with 2,000-10,000 images
https://github.com/fragofer/voctree
http://www.ipol.im/pub/art/2018/199/

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.