The Computer Vision has made impressive progress in learning to identify leaf mines since I started identifying them on iNat a few years ago. But lately I’ve noticed that it’s been going a bit too far: observations of plants that happen to be hosts of commonly observed leafminers are getting identified as the leafminer species, even when there are no leaf mines at all in the photos. In some cases the CV isn’t even giving the plant ID as a suggestion anymore. Some examples include photos of Arctium being identified as Liriomyza arctii, Ilex opaca being identified as Phytomyza ilicicola, Impatiens capensis identified as Phytoliriomyza melampyga, and thistles as Phytomyza spinaciae. This problem probably began with an abundance of observations that aren’t adequately cropped, showing a tiny leaf mine in a sea of unmined leaves. I’m sure it is exacerbated by people agreeing with these misidentifications so that they become “Research Grade.” I’ve never understood what motivates people to go around agreeing with IDs seemingly at random, but I think it must have to do with the leaderboard. It would be an interesting experiment to try removing the leaderboard* for a while and see how it affects the accuracy of RG observations, as well as the accuracy of CV suggestions. Failing that, maybe it would help to add some kind of prompt that asks observers whether the observation is for the plant or for some other organism that is on/in the plant. Forcing users to answer this one little question before blindly accepting the top CV suggestion could go a long way in improving the quality of IDs.
(*) I would love it if there were some way to hide myself from the leaderboard. Because I have identified thousands of observations of leaf mines made by agromyzid flies, for example, I show up as one of the top identifiers of Diptera, so I get tagged all the time on observations of all kinds of flies that I know nothing about.
That’s good, but the problem I am referring to is caused by users who are evidently not thinking about what they’re doing at all, so this advice doesn’t apply. I think the users making the initial wrong IDs are maybe mostly using the Seek app? I’ve never used that, but am I right in thinking that it automatically selects the CV’s top suggestion and posts the observation with that as the ID? And similarly, this solution doesn’t address the issue of people going around and agreeing with clearly wrong IDs.
Since there is no indication whether a suggestion is for a plant or an insect some observers may just believe the first suggestion is for the plant they see. For many galls and leafminers there is simply no common name in most of the languages and binomens say nothing to the casual observer.
Yes, that’s what I was trying to address with my suggestion of a prompt asking the observer whether they’re trying to ID the plant or something else in the photo.
I don’t know how the CV is penalized when it gets wrong and the thing it got wrong turns out to be in the “other side of the Life spectrum”. In other words. You don’t see the Similar Taxa for a leaf mine or plant to be the plant taxon or the leaf mine that people mis-ID.
I read somewhere that once a species get 50 or so research grade obs the CV starts tracking the mistakes and builds the similar taxa thing based on ID error and fixes we do. Maybe we need a system that could give the two options based on a first review of the CV like:
CV: Oh so I’m confident you have here a maple tree, so maybe you want a leafblotch miner or you want the tree only. I’ll give you some pop-up thingie for you to choose from host options to parasite options since both are valid for this case.
Also, I see many incorrect leafmine IDs from north hemisphere in South America too. There should be a bunch of warning signs before they let people pick or agree to IDs that are very far from the distribution range. Sadly the leafmine knowledge down here is very scarce compared to yours, so our only way to fix things is knocking it back to Pterygota or something.
You can see on the observation how it was uploaded (Seek, iNat Android, iNat iPhone, web). But Seek us using an older version of the CV, so if this is a new problem I don’t think it would be from Seek observations.
Okay, I just looked at the most recent Phytomyza ilicicola observations and see that this one is a CV suggestion, uploaded using the iPhone app: https://www.inaturalist.org/observations/235710307
So no particular reason to blame Seek. Note that this is just a picture of Ilex opaca leaves, and the only CV suggestions are P. ilicicola and P. opacae.
And here’s a recent “Liriomyza arctii” observation that just shows Arctium leaves, again a CV suggestion using the iPhone app: https://www.inaturalist.org/observations/232577311
The first two CV suggestions are leafminers that use Arctium as a host, and the next six are a random assortment of insects that aren’t associated with Arctium.
I suspect that agreements that seem mindless are less about getting on leaderboards than about truly believing that the iNaturalist ID must be correct and agreeing because that seems the right thing to do. (I did that myself, when I started and had no idea about leaderboards at all.)
This is because an insect has already been selected as an ID, so the CV is now only showing insect suggestions, as explained in @thebeachcomber’s screenshots above. If someone disagrees with the ID, bumping the observation back to Life, the CV will start showing all taxa.
I think Thomas’ original point was that you only see insect suggestions because there is already an ID of the insect. If you started with no ID you would probably get plant suggestions. I actually do get plant suggestions even with the insect ID:
Yes, I forgot to apply that new knowledge here, sorry! It must still be the case, though, that the CV is initially giving the leafminer as the top suggestion, otherwise people would be choosing the plant.
Strange that it’s giving different suggestions to different people–or else different suggestions to phone app (you) vs. desktop (me).
So far people’s attempts to do that have mosty resulted in confusion and misidentifications. Names like “aspen leafminer” don’t work because any given plant typically has numerous leafminers that feed on it. “Locust digitate leafminer moth” is a rare example of one that is actually useful. In any case, common names don’t address the problem I describe in this forum post, because the first three of the four specific examples I cited are leafminers that have common names on iNat: “Burdock Leafminer” (which isn’t a good one because there are multiple burdock leafminers), “Native Holly Leafminer” (same problem), “Jewelweed Leaf-miner Fly” (this one is reasonable).
