Computer vision clean-up (archive)

I would recommend adding Flavoparmelia caperata, Lepraria finkii, and Chrysothrix candelaris to the fungi list. These are all common species that seem to be applied to anything that looks vaguely similar.


I would like to add Mercurialis perennis (a European plant) but I’m not sure about its potential distribution in Northern America. The Computer vision often suggest this plant for bushy green plants. Before adding it to the list, does anyone know whether it has been actually introduced to Northern America ? I could not find anything about it (my sources are quite limited).

USDA Plants does not list Mercurialis perennis.

I added a “Common Themes” section. Please add ideas there!

Looks like there’s just a handful of observations of Mercurialis perennis in North America remaining anyway (I reIDed around 30).

I corrected many observations, especially the ones in tropical or desert like locations. I’m still not that sure about the ones in North America. There’s one particular observation that makes me wonder

added Annelida section and specifically Lumbricus terrestris. Please add any worms you see, and keep this thread alive.

1 Like

I would like to help, but I don’t know any of these species. If anyone would consider doing a short guide like those here, that would be fabulous.

Many of the flags do have users providing ID tips. There’s not a great place / format for this type of information to be hosted, but I’d encourage people to add to the flags for now.

( see also: )

As a non-expert this looks like a great way to help out, and I’d love to contribute! However, I live in Australia and I think I would be more useful with species that occur in Australia. Most of these are North American so I’ve never seen them in person and I’m not familiar with the species they might be mistaken for, and besides Australia seems to have fewer identifiers in general so help seems to be more needed :)

I wouldn’t know enough to do it on my own, but I’d love it if some experts could add suggestions of taxa that need cleanup within Australia. I’m sure we have a few problematic species where the AI tends to suggest a globally widespread species instead of a local one that has fewer iNat records. I searched the forum but I didn’t see any separate threads for Australian identification help - please let me know if a resource like that already exists.

One thing is that while there are a lot of North American species here, there are probably a lot of incorrect observations of them showing up in Australia.

Since Australia is one of the somewhat newer communities on iNat I’m not sure if there are a lot of computer vision errors of Australian taxa yet.

Salix amygdaloides is another species the computer is over-enthusiastic about. This north temperate species native to North America was reported from Europe, Hong Kong, Singapore, Malaysia, Mexico, and California. Fortunately, all but one of those records were easy to get rid of. (I suggested an alternative for the one challenging one, a closely related species that’s in range.) There are errors in North America, too, especially in the southeast, but I’m going on to other things. There aren’t enough out of range S. amygdaloides records to add this one to the list, but I though I’d let you know about it.

1 Like

I’ve added Lasthenia to the list - Computer Vision seems to suggest it for any yellow asteraceae, so there’s tons of misidentified observations.


CV is also pretty clueless when identifying blue/purple asteraceae. It is very good in realizing that your subject is the flower - anything from a full frame of the capitulum to a blur in one corner to pale dots in a meadow and it knows what you were looking at - but beyond that it just seems to pick whichever species is currently at the top of its list and apply it worldwide. However, it does not seem to have a single species that it always chooses.

I cleaned up a couple of hundred Aster alpinus in North America a few months ago, and have worked on a couple of other species. Any common “aster” (sensu lato) or Erigeron is probably a good candidate to have lots of questionable CV suggestions.


Yes, I think asters in general are quite a mess. There’s just so many of them, and often very subtle differences - I keep telling myself I’m going to learn them, and then getting intimidated hah.


Another one is virtually every photo of a carnivore track in snow is suggested as a Fisher, which given their relative scarcity over much of their range seems improbable.


1 Like

Added Taraxacum to the list. They’re difficult to ID to species at the best of times (POTWO says Taraxacum “Includes 2343 Accepted Species” - yikes), and it seems to be the default choice for almost every yellow flower - it’s a mess.

I don’t know enough to differentiate within the genus, but I’ve been working on weeding out the asteraceae that definitely are NOT taraxacum.


Taraxacum, oh yeah:


yeah, i wonder if there is a better way to track dandelions with some subgenus type division for that complex of European ones that occur throughout the world, such as red seeded and officinale. I don’t personally think trying to use iNat to map microspecies is a good idea, but there’s more about that in that thread you linked. In the mean time, yes, proceed with caution with Taraxacum.

Another one on the east coast of North America: nearly all Toxicodendron listed as radicans when rydbergii is often more common. And to make it more confusing some in Canada appear to be using a flora that combines them, but since iNat doesn’t, they get lumped into radicans when they do not. Any time I see an observation of T. radicans I just assume it is one of those two.

1 Like

Another disturbing one, though one that affects mainly “casual” observations, is that for almost any double-flowered rose, AI will suggest “Rosa chinensis”. Further, casual observers will tend to accept that without question and then others will agree with it. This seems to be encouraging a false belief among observers that the complex hybrid roses in cultivation would be correctly described as Rosa chinensis. How about tweaking the AI photo recognition to just leave it at the genus level, Rosa?