Out-of-taxon suggestions from visually-similar search with taxonomic restriction

Recently, I’ve now-and-then been seeing some rather crazy results from visually-similar suggestions on the ID screen. Searching within a particular plant genus, the top suggestions are sometimes animals, and I can’t find any similarity in the taxon names that might lead to confusion. Today’s example while IDing https://www.inaturalist.org/observations/27582909 gave three mammals, two birds, four unrelated plants, and one insect as the top suggestions for a search within the taxon Balsamorhiza (a plant genus). It seems to be pulling nearby species regardless of visual or taxonomic similarity. If I change the taxon selection to the next higher category (Tribe Helianteae) the search works as it should.

2 Likes

“It seems to be pulling nearby species regardless of visual or taxonomic similarity.”

I wonder if it’s really pulling nearby, because I’ve noticed that “visually similar” under the compare tool usually ignores location, but it makes it obvious by removing the field for the location filter.

I wonder if “visually similar” is meant to ignore taxon too, but is not hiding that field.

I could sort of see the logic if it is meant to ignore all other filters.

Along a similar thought, I’ve wondered why pictures of humans are often suggested to be Agaricus bisporus (button mushrooms). I think it is because one of the few research grade observations of the species (most are grocery store specimens or other Agaricus species) includes several pictures of the (human) identifier. I wonder if the propensity of people to pose with their giant puffballs (Calvatia gigantea) makes that another common suggestion for humans.

1 Like

I also think it’s because many deer photos have leaves and maybe fawn hidden in leaves would look the same for program.

1 Like

No, “visually similar” does normally respect the taxon selection. As I said, it works properly (showing only plants within the taxon) if I select a higher-level taxon for the search.

I have been having this issue as well. I’ve confirmed it happens across multiple taxa. After some additional playing around, my current theory is that it happens when there is no “visually similar” match in the selected taxa. For some reason instead of displaying “no results” it instead grabs the list of the most commonly observed taxa in the immediate area and ignores the filters.

Take this pelican, for example: (visually similar) + (taxon: pelicans) comes up with a couple of reasonable matches.

However, when I ask it to find a (visually similar) + (Taxon: bears) it obviously can’t find a bear that looks like this pelican so it spits out this:

The suggested species in the second case are all in the top “most observed” list for the county:

It may just be a case of the system pulling in the wrong list instead of showing a null result.

Have you noticed any searches using “visually similar” filter returning a “no suggestions” result? I thought I had, but going back to double check I only seem to be getting that result when I try it with extinct species:

1 Like

On yet another thread they said iNat image recognition isn’t saying - oh look a bird, wonder if it’s a pelican. It is purely picking up the shapes and colours … and last time you said that was a pelican / fence lizard / field of greyish daisies.

If you’re using the Identify page, I wonder if the modal hasn’t “caught up” to the current observation and is using info from the last one. Just now I identified something as a Bumble bee, went to another observation (with no ID) and clicked Suggestions. When I switched to Visually similar, the taxon restriction immediately went to bumble bees:

2 Likes

I was able to replicate by going to the observation linked in the first post (https://www.inaturalist.org/observations/27582909), clicking the “Compare” button on the ID, switching the source to “Visually Similar” and changing the taxon to genus Balsamorhiza.

1 Like

Nope, certainly not that. This arose when I had been on the observation for a while, and was narrowing down the choices. Things went crazy only when I got down to genus level.

1 Like

No, that’s not it. You can make this happen outside of the identify page, by going to an observation and clicking “compare.” Choose visually similar and then choose a genus that has no similar enough observations. Expected behavior would be for it to then just say “no matches” but instead it spews out a list of species found in the area, which are NOT within the genus the filter is set to - and not at all visually related either.

1 Like

[quote="

Just to clarify, the issue is not that the taxon restriction is changing (I’ve experienced what you describe there as well) but that the suggestions are not within the selected taxon at all.

3 Likes

same here, sorry for reposting but I placed in a wrong thread.
Here is showing bird suggestions after filtering for Thricoplusia so I bet taxon filter has got some issues?
I have never willingly tried to identify birds so far…

1 Like

Another example, this time at the section (sub-genus) level. For observation https://www.inaturalist.org/observations/40680691 of Western Juniper, clicking “Compare” then source “Visually Similar” goes straight to the section level (“Scale-leaf junipers”) taxon and suggests a mix of mammals, birds, and pine trees, but no Junipers:


If I edit the taxon selection up one level to the genus (“Junipers”) everything is OK, and the correct species comes up as the second suggestion:

1 Like