Step 2: Identify Page
(Filters: e.g. Taxon = Silene, Place = Norway, Date 2025-06-21)
Step 3: Suggestion-button
Step 4: Two different taxa pops up, but they have the same first picture.This picture is not found in either taxa’s selected taxon-pictures. The second pictures in the suggestions are the respective priority picture. (I have not found the “problematic picture”, but my guess is that it belongs to the second suggestion, S. vulgaris.
The same happens when using CV from the Explore page for this observation.
Interesting! From the original observation pointed to by @ShotShot, S. uniflora ssp. uniflora seems to be the correct id.
This raises a few questions (to me). I haven’t studied the inner workings of the CV, merely registered that it’s suggestions seems to get continously better (for many taxongroups). And I’ve noticed the very useful new feature of presenting the “most similar” picture. However, this example makes me wonder how the data for training the CV are selected? Is this a dynamic process or is it the set of pictures id’ed as a given taxon at a given time? And why does CV present at non-FG obs as most-similar (the obs in question here is not RG as it is informal - planted, - Id-s can of course be correct eventhough the plant is planted, but among planted ones many are horticultural variants)?
As this picture is presented for two taxa, has this pic been included in the training material for both taxa? This would be problematic, I think.
The taxonswap in question was done nearly 9 years ago.
No, if used at all it will have been used only for the active taxon when the export was done earlier this year. Anyone can go in and remove the photo from the inactive taxon page. It’s just an illustration and its appearance there has nothing to do with the training of the computer vision.
Cultivated plants are included in the computer vision training as long as they have a Community ID (minimum of two agreements as to species).
Which would be wrong*: the photo is definitely of S. vulgaris ssp. maritima – “taxon change” at iNaturalist or not, “inactivation” at iNaturalist or not, accepted synonymisation with S. uniflora or not.
*or else, it would certainly be wise and consistent to remove all photos from all inactive taxa, as soon as they get inactivated. Although it could also complicate matters, whenever people rely on visuals to understand what taxa were dealt with during taxon_changes at iNat.
Fair enough. Another option would be to make a feature request so that taxon photos from inactive taxa no longer appear anywhere else. That would be a better solution, but will take time, if it happens at all.
Ok, that makes sense, of course that the data used for a given taxon only includes pictures identified to that taxon at the export moment.
What remains as confusing is the algorithm that selects the “most similar” picture. Shouldn’t that search be restrained to the current understanding of the taxon? I.e. such that the pic in question should not be presented for the Silene vulgaris alternative (in this obs) as this picture does not belong to the S. v. taxon (according to community ID).
I think that’s a valid point and could be the basis for a relatively straightforward feature request, if you or someone else wants to put one forward.
Description of problem: The same pictures shows for two species. This was a split not too long ago and this might stem from that. However I can only find the picture for borealis
