I am just trying Digikam on a very large collections (140k).
The file hierarchy has several symbolic links that point to directories to help the navigation etc.
Unfortunately, the discovery process follows symlinks (doesn't resolve path), thus it makes zillion of duplicates.
In my case, I have some top links, so the collection is basically multiplied by two, three, ...
So Digikam is useless on this case.

Solution:
- implement a fix to the discovery process
- and implement a cleanup process for the database
- or add a configuration option