CV suggestions are bizarre, not "visually similar" at all

[drbh2o]

Ah yes, the feared pengueevil.

[tiwane]

@raphetnature

Here stange results I had yesterday for a beetle photo.

You entered the date and location before getting a suggestion, correct?

@sunguramy

I noticed more weirdness myself lately. I mean whenever i have “gills” in the file name for a mushroom (i tend to name images mushroom 01 - gills, mushroom 01 - stipe, etc) and when i combine those it still sees gill and wants me to call it a Bluegill and suggests all fish. If I have turkeytail in the name, it suggests wild turkey and a bunch of other birds. I assumed it was due to naming convention i use for files.

That based just on file names (or keywords in the photo’s metdata), not computer vision.

@sunguramy

Was convinced my earthball was a rose; for example. The CV seems to assume bird or garden plant (not wild plants but showy planted/cultivated flowers) for my mushrooms as of late. Even with location and all files having mushroom in the name.

We’d need the exact photo (including metadata) and a screenshot of what you’re seeing. Note that posting images to Discourse removes metadata, so you can email help@inat or share a Dropbox/Google Drive/etc link here.

[sunguramy]

@tiwane

That based just on file names (or keywords in the photo’s metdata), not computer vision.

Aye, as I said I figured it was due to the file name so I always ignored it. When I clear the auto fill it would come back with better.

If I run into the new issue again, I will do that.

[alex]

Would you mind sharing the original photo? You could either post it to iNaturalist and then share a link here, or email the photo to help@inaturalist.org. I would love to look a little closer to see what the CV model thinks about this photo.

Thanks,
alex

[becksnyc]

I did an upload today and got bizarre results. An octopus brought up “human,” no suggestions at all for very clear photos of manta rays, sea anemones gave suggestion of willows (the plant). Highly unusual, as my photos are all excellent quality. I had to find and choose the category myself for most of my submissions.
I just re-uploaded the same photos to take screenshots and everything was working normally.

[alex]

Hi Becksync,

Screenshots would be great when you run into this!

Thanks,
alex

[GothHobbit]

I found a couple more.

The second one makes an odd kind of sense because ‘beaver’ is part of the photo title and location data, but surf clam for a Twelve-spotted Skimmer …I got nothin’.

[alex]

can you also share the observations or the original photos? thanks!

[thebeachcomber]

Were those actually suggestions offered by the CV, i.e., you manually clicked on/selected those IDs from the dropdown? Or are you referring to this behaviour: https://forum.inaturalist.org/t/add-observation-automatically-tagged-my-observations-with-incorrect-names-based-on-photo-filenames/8203

[GothHobbit]

@thebeachcomber, both were suggested by the CV; all I did was move the cursor to the name field. The second example has ‘beaver’ as part of the photo file name (location). The first one of the Twelve-spotted Skimmer coming up as a clam …it’s not the first time that has happened, and I know that the CV isn’t pulling it out of the file name or metadata on that one.

@alex, be happy to! (In fact, I just got another one where the CV model appears to be yanking a suggestion from the file name.)

[DianaStuder]

@GothHobbit

CV model appears to be yanking a suggestion from the file name

That is working as intended. You have to find a way to work with, or around, that.

[GothHobbit]

Twelve-spotted skimmer observation

This is the most recently posted observation where both photos came up as Atlantic Surf Clam. Here’s one of the original photos.

@DianaStuder, okay, I wasn’t sure about that. I thought that might be the case when there was something in the photo name to suggest it. The dragonfly, though, is a mystery.

[jonasgruska]

Received very strange suggestions for this lichen observation. Did something change with the algorithm? I am used to the observations being pretty good most of the time.

https://www.inaturalist.org/observations/145141065

[fffffffff]

System says it’s not confident, when it says so most likely its suggestions are wrong, it can’t recognize the patter on photo.

[DianaStuder]

If you toggle off Seen Nearby, there is a species suggestion which seems possible.
(but I am limited to … it’s a lichen?)
CV is determined to try hard, and we have to decide if we agree.

[jonasgruska]

You are right, turning off the seen nearby helps! I was just stumped how off the suggestions were. Yes, it is a lichen.

[bernhard_hiller]

Among the funniest identifications I’ve seen yet: a beginner photographed his dog, and identified it by CV as Sus scrofa (wild boar):
https://www.inaturalist.org/observations/145193571

and cattle were identified as dog:
https://www.inaturalist.org/observations/145193359

[DianaStuder]

Perhaps the observer was battling with using iNat in English (since I in turn find the Thai script beyond daunting!)

[fffffffff]

No, it’s normal, for my mammal observations cv can suggest anything, dogs are really hooved animals for the cv. Look what it suggest already RG obs of a dog (so ids affect cv)
fox, deer, maybe a dog or a squirrel?

you ided it as a dog, but what about a goat?

[DianaStuder]

then I would pick the vicuna