Opinion on the computer vision icon?

Agreed. We all should get past our egotistical selves and use what tools are available. Amateurs get it wrong sometimes. So do experts. No shame in that.


Sometimes I use it to jog my memory. I see an organism, and I think, “I know that one, what was the name?” I look at the CV suggestion and think, “Oh, yes, that was it.”


Agreed. I will often use the CV in the Identify Compare modal to replace my typing. I’ll switch to “Visually Similar” as a “Source” to verify distribution sometimes zooming out on maps to find out if there are likely other possibilities of dopplegängers and even use the GBIF network overlay to see if there are holes in the iNat netted observations. Sometimes this will help me realize I need to switch to subgenus, genus, or family.

Use it as a tool and not a rule…and check those notifications for disagreements…and check for mavericks.


Sad to see this paper getting reproduced in print media now too with the bad data remaining


I also think that it creates lots and lots mis ids; I almost never use it.

quite a useless example of AI

That leads to a dumpyard (big amount) of mis ids.

Are you aware that CV adds new species every month?
iNat staff prefer to use the term Computer Vision, since it is not artificial intelligence.

23 insect species from India added from 19 February data

And for Mr Fizzy in the Netherlands
18 insect species from the same February based update

End of April ish we can expect the March based update.

I wonder why this thread started by favouring I Can TYPing over a quicker more efficient work flow. The CV is a suggestion, it is still up to you to evaluate - is it right, did I get the taxon I expected to (or a mis-click instead?)


I think that computer vision models should look for features; not just assume; like there is a penthicodes warleti, then it will id as atomaria;

Instead (as an example) it should suggest genus penthicodes and if spots are visible then it should evaluate (atomaria has 2 spots along the margin of tegmina, but warleti has 2 spots in the centre of the tegmina)

I must be daft, I have never noticed a CV icon?

I had to go look for it. I found it. I thought it was a first ID symbol!

Ive never been able to type out a whole name upon upload Im not that great a speller. My uploads probably ALL have that. Even the ones it auto IDd to species from file name it looks like. Even for example this one of mine (my own obs i am linking here), I used the CV to get me close, wasnt sure on species it was suggesting, went to taxonomy tab to compare, then copied the genus and then went back to my upload window and pasted the genus name is as text. It always pops up that window below and I click it. If I dont, it doesnt like the name and wont save what I typed in. Since I pasted in the name as text, I didnt click a suggestion until after pasting in the
ENTIRE correct spelling…yeah. Ha! If that popup as one types that you have to click gives a CV symbol, lolllllll all mine are gonna have a CV symbol because thats what it did there. Its what I always do!

So…Personally its a worthless symbol. Ive apparently AWAYS used cv (apparently!) even when I paste in the spelling from my file name so it saves right and doesn’t upload without ID! As again, if I dont press that popup window suggestion, it wont save my ID at all.

Which leads to a question: should this be a bug report??


That… might be a bug? Or it might just be wonky? Personally I really don’t care about the icon unless I notice something egregious going on.

I actually ran into an observation that had the opposite problem the other day; someone had typed in ‘Black Gum’ as the common name for a tree in Ohio, they obviously new what it was, but the plant that popped up and they entered was Eucalyptus aggregata, which is an Australia endemic, and not Nyssa sylvatica, which iNat lists as ‘Black Tupelo’ but which I’ve only ever heard Ohioans call a Black Gum.

Though when you do type it in, it comes up as Black Tupelo (Black Gum) - I can only think this user skipped past the two Nyssa species that had Black Gum in parentheses and went to the entry that just said Black Gum, which TBH is a totally reasonable mistake to make.

Tl;DR sometimes its wonky even when you do type it in


Just typing in and using the autocomplete doesn’t mean the CV was used, so I think that you’re describing a separate issue. If you type in all or part of a name before the CV model output arrives, or paste in a name, it doesn’t give a CV icon to your ID - just tested on my own observations to be sure.

If you type in anything, iNat will give you the closest match taxa to your text (just like it does if you are IDing other people’s observations and not using CV at all).

1 Like

True. Yeah, its more of a weird thing that happens when you half type something in and maybe not paying enough attention to the auto complete.

1 Like

What an interesting thread. I’m with Sunguramy - I had no idea this was even “a thing”. I knew nothing about any icon and I’ve used the CV for every upload I’ve done.

I have what is probably a “curious” methodology. I know many of the organisms in my area but for those I do not, I use the CV to get suggestions. I then double check things, visit other websites. Once satisfied, I remove the photos, add names to my files etc, and re-upload and use the CV to add the names.

For the organisms I know well, I also just accept the CV when it’s correct.

I have moderate chemo-induced neuropathy and I’m not going to type any more than I absolutely have to.

So while there might be some utility in this icon (knowing where an ID came from), without knowing “the rest of the story” about how users are using the CV, the icon itself appears to be of little value.

BTW - about that Nyssa. Here in NC, and coastal SC, many of us call it it Black Gum.


Do you have any idea how programms like that are working? It’s not what you describe and you’d need a completely different sci-fi software to get this, esp for thousands and thousands of species.

1 Like

or, a human, taxon specialist. Like @fffffffff for insects.

In this paper https://arxiv.org/pdf/2203.03253v1.pdf 91.39% is mentioned.

Compared to existing published works, our method
consistently achieves SOTA results on multiple datasets,
specifically 76.81%, 83.67%, and 91.39% top-1 accuracy
on iNaturalist 2017, 2018, and 2021, respectively


1 Like

I wonder what 2022 would show? Still climbing gently?

I hope so.

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