Alternatives to identification help when computer vision says "We're not confident"

General
1

I’m sure everyone has gotten that message at some point when asking for an ID. So, any next steps?
I recently ran into a case where iNaturalist couldn’t confidently ID an admittedly so so image of a moth, so … I ran it through Google image search. Just to be clear, this DOESN’T ID anything, but can often find similar images on websites that might actually have a solid ID! Google did in fact find four similar images, one from a solid website and after doing a little more research, I had to agree with the ID.
So, iNaturalist is certainly my first attempt, but maybe try image search if you don’t get a useful answer.
Just an idea.

6 Likes
2

It helps for commonorgarden exotics.

1 Like
3

I also use google image search to look for similar things. It is useful because you can tell it which part of the photo you are interested in, or use a photo other than the first one in the observation.

1 Like
4

I use this method with more esoteric looking things to identify potential orders or families it might be in to further narrow my search or just to leave as notes for other identifiers (mainly first link). Here’s three really weird / interest things I found

1 Like
6

I wasn’t sure what the topic title was referring to, so I added a bit of text to it. Please feel free to revise.

1 Like
7

I might be old school, but i still think it’s best to ask the informed people who are freely giving their opinion on identification of very similar things!
(and here, there’s often a community of top identifiers who can be contacted)
I find the automated systems of various websites can give confident answers often based on very little that’s secure - such as sites where a name is tagged on by someone who actually had no clue! If here CV is saying “We’re not confident” then great - unless your creature is common and really easily diagnosable, that’s the common reality for much of biodiversity!

7 Likes
8

If something is an “unknown” - you can’t ask an informed person until you can get it to at least family or so. You don’t know which informed person to ask.

1 Like
9

But you do know it’s a moth or aster or grass so you can identify it at the level you are confident and wait for somebody to ID it.

3 Likes
10

If all you’re confident of is that it’s a dicot, though, the chances of an expert finding it feel increasingly slim, due to the growing backlog.

5 Likes
11

I do that regularly. But occasionally I find something that seems quite distinctive but I don’t know what it is and the CV can’t recognise it. Then a google image search can give me clues of where to look.

12

" Then a google image search can give me clues of where to look"
and equally perhaps clues about which of many informed people to try to ask?

Even with the most broad taxon groups “moth, grass, etc” the system here provides details of potential identifiers, who can be restricted by geographic region of focus etc. If you can’t work out your six legged arthropod is at least an insect (which might be a defensible starting place for a potential “moth”), then maybe trusting the automated systems on the far more precise names they tend to feely push users towards isn’t wise.

13

My personal ID philosophy for dicots is somewhat more lenient than for other high-level taxa. Because I know that putting something in dicots is very unlikely to nudge it towards an ID in the next few years, I’m willing to apply an ID at family or order level with a lot less confidence than I would do, say, for a monocot. Maybe CV isn’t much help, but if I can see some feature that looks plausibly like a plant family that might occur in the region, I’ll often go for that.

My thinking is, this way, there’s a chance someone knowledgeable will see this and refine or refute the ID. And I’m still avoiding a precise ID that risks creating an unreliable RG observations when the observer uses “Agree” as a Like button.

5 Likes
14

If I can get something that was “unknown” to family or genus, then the expert may well find it without being annoyed by tags from me.

One thing I use google image for is cultivated plants. The CNC creates massive amounts of cultivated plant observations. The iNat CV is not great at IDing cultivated plants, and expert identifiers probably aren’t interested in IDing cultivated plants.

Another time it is useful is for species which the iNat CV doesn’t know about due to small numbers of observations - particularly invertebrates. Google can often turn up something that at least looks similar.

And in reply to your last sentence, I do not “trust automated systems” the way you are implying. I use them as an aid, and always do further searches before adding an ID. (I just did a quick check - so far today I have added 238 IDs, and I used google lens on 1 of them. And in that case I was sure it was a particular species, but the observer had IDed it as something completely different. iNat CV couldn’t ID it, but google agreed with me, so I added the ID I was intending to.)

3 Likes
15

Same here, and once I’ve got some leads I redo a search with candidate trade names or the common name and “for sale” added. That usually brings up plenty of pictures of cultivated/garden plants to compare against. I check the list of websites returned for nursery/garden center websites. If it only pops up on Ebay/Instagram/Amazon listings etc. but no reputable gardening sites it may be a made-up name. If I can narrow it down to specific cultivars, I check cultivar names with “origin” or “breeding” or “hybrid” added to search for evidence of hybrid breeding, in which case a species ID is not appropriate (I add a note “garden hybrid cultivar” in those cases and mark them cultivated).

4 Likes
16

There are countless resources online- conservation department websites, region checklists, independent hobbyists, etc. Just look up “moths of south carolina” or whatever, and I bet you can find what you’re looking for.
Side rant: This is example #100000 of why I think AI should not be the automatic first line of defense in identification. We’re forgetting how to do casual research. Even by doing a reverse image search, I think you are cheating yourself out of the opportunity to build yourself a network of resources.

2 Likes
17

True enough, but there are also variations on this commonly seen copypasta:

It is helpful to add your own ID at the narrowest level you feel comfortable with, for example, “plants” or “beetles”. Observations with an ID are much more likely to be identified. If you leave your observations as “unknown,” rarely anyone will look at them.

I find that with plant pathogens, for instance, using Visual Search can help me get at least an initial ID – right Kingdom, maybe right Family.

1 Like
18

I’ve also found Facebook groups to be a good alternative to getting some community sourced ID ideas when there isn’t much activity on my observations in iNat.
There are lots of groups on Facebook for different taxon (some groups even dedicated to singular genera). I usually post my original pictures in the group (I’m mostly doing this for fungi), and add a link to my iNat observation in the comments. It’s a hit or miss, but I’ve learned a lot with this method!

1 Like
19

for example.. here are some of the facebook groups useful for identifying fungi
https://www.facebook.com/groups/426474384203464
https://www.facebook.com/groups/117808248330980
https://www.facebook.com/groups/1510123272580859
https://www.facebook.com/groups/1117745858759221
and moths:
https://www.facebook.com/groups/137219092972521

there’s lots more groups for different taxa out there, but not all are equally active

20

I get what you mean, but I personally believe that computer aided image searches can make an amazing starting point for precisely the casual research you mention. The important thing is to remember that they are essentially just that… a starting point. When I come across an organism that puzzles me and I just don’t know where to start looking, I’ll sift through my memory for images, then I’ll maybe ask my husband who’ll sift through his memories, or a friend. We’re effectively doing a human aided image search. Consulting iNaturalist’s computer vision, or Google, is in my view just a way of widening that image search exponentially. And from there (hopefully) comes the inspiration I can then follow up to elaborate my best shot ID.
In my mind, the problem is not the initial image search, but the fact that many people will just stop there, without even realising that more research is needed. I see this as a matter of educating people, rather than a problem with computer-aided image searches themselves.

3 Likes
21

The other day another iNat user reported that he was banned in a similar group for mentioning iNat.