Peculiar results from Computer Vision - top suggestions have neither taxon image nor observations

I’m working on “restoring” Campanula rotundifolia observartions in my area after all observations were lumped into the Complex Campanula rotundifolia. When using CV from the Identify-page in the web-application, the two top suggestions are C. nejceffii and C. ruscinonensis. These two members of the large C-r-complex have no observations and no taxon photo. As I understand CV this system is trained on pictures from iNat and it seems peculiar that these taxa should be suggested.

In addition CV returns the same ten taxa guesses as I go through observations and these ten follow in the same order - C. rotundifolia s.l. always the last one. Another strange thing is that there are only two taxa that have any observations in the geographical area of the observations so CV does not seem to take geographical information into consideration.

All this is very surprising as I find that CV has become increasingly good at suggesting taxa - for many flowering plants at least. Can this “bug” have something to do with the C-r-complex of 35 subordinate species?

5 Likes

Hi, wecome to the forum. URLs/Links and screenshots are helpful to communicate an issue you are encountering.

e.g.:

https://www.inaturalist.org/observations/identify?taxon_id=984576&place_id=7016
https://www.inaturalist.org/observations/identify?id=200771116

2 Likes

Thanks for this. Your example show the issue. What I most often get is this:

2 Likes