Will a single identifier's efforts signifiacntly influence the AI identifications?

About a month ago I noticed that a large quantity of identifications for genus Hydrangea were incorrect- many many garden cultivars of mophead hydrangea (Hydrangea macrophylla) and panicle hydrangea (Hydrangea paniculata) were mis-identified as wild hydrangea (Hydrangea arborescens), even to research grade. This is understandable, as there are a plethora of horticultural varieties of Hydrangea growing all over the temperate zones, and cultivars of the same species can look quite different. Even ‘wild’ hydrangea has many cultivars, and so an observation of a wild hydrangea is as likely to be an intentionally-planted garden variety as it is a truly wild organism. Another major cause of this issue is the auto-ident tool- it needs refining for this genus. Unless the observation shows a big blue ball of a flower (mophead hydrangea), the AI is most likely to recommend ‘Wild Hydrangea’. Bit of a mess, really, but also kind of satisfying to try to sort out.

To the question- I’ve been enjoying combing through the ‘wild hydrangea’ taxon and clearing up mis-idents, all the while hoping that if recommend the correct ID to a panicle hydrangea that’s been mislabeled as wild hydrangea enough times, the AI will eventually recognize the differences between wild and panicle hydrangea. Will my efforts alone move this needle, or does the AI require the input of multiple reviewers to adjust itself?

Thanks all!


There is no impact at all on the AI about the number of people who identify or review a species.

It used to be that to be considered for inclusion in the training set, the photos used had to have been submitted by a diversity of users. I’m not sure if that is still in place.

The next version of training the model should definitely be helped by your corrections of misidentifications! However, keep in mind that it’s only updated once or twice a year at this point. Thanks for all your efforts.


I have noticed the effect you are asking about. It happens especially when a reviewer is the only identifier of a species per given area. I observed it before, and still notice it with Iochromia fuchsioides is being suggested to be fuchsia boliviana.

Ah, I didn’t realize the adjustments were only a few times a year. Good to know-thanks! I’ll keep at it.

1 Like

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.