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

[suecar]

Please fill out the following sections to the best of your ability, it will help us investigate bugs if we have this information at the outset. Screenshots are especially helpful, so please provide those if you can.

Platform (Android, iOS, Website): Website

App version number, if a mobile app issue (shown under Settings or About):

Browser, if a website issue (Firefox, Chrome, etc) : Safari

URLs (aka web addresses) of any relevant observations or pages: https://www.inaturalist.org/observations/upload
https://www.inaturalist.org/observations/144608140 [a observation I manually IDed and subsequently tried “Suggest an Identification”]

Screenshots of what you are seeing (instructions for taking a screenshot on computers and mobile devices: https://www.take-a-screenshot.org/):

Description of problem (please provide a set of steps we can use to replicate the issue, and make as many as you need.): I have been uploading a set of moth photos. For some observations, the ID suggestions are appropriate, but about half the time I get a bizarre combo of suggestions, none of which are similar at all, and many of which are not lepidoptera but instead plants, seashells, birds, etc. [see first & second screenshots]. Interestingly, if I enter an ID manually for an observation that received bizarre suggestions, and then later I click on Suggest an Identification for this observation, the suggestions are more appropriate, though still not great [see third screenshot].

Step 1: Add photo to upload page (https://www.inaturalist.org/observations/upload)

Step 2: Add location to observation

Step 3: Click in “Species name” box and wait for list of suggestions to populate

[tiwane]

When I put the image in the uploader (and add the date and location) I get basically the same suggestions you do, although slightly different:

If I remove the “seen nearby” restriction, I get a better top result:

Sometimes you’ll want to remove that restriction, as there could be visually similar species to what you uploaded, but none have been posted to iNat nearby. I don’t think this is a bug.

@suecar

Interestingly, if I enter an ID manually for an observation that received bizarre suggestions, and then later I click on Suggest an Identification for this observation, the suggestions are more appropriate, though still not great [see third screenshot].

Yes, the community ID is taken into account when suggestions are given.

[tiwane]

(Also changed “AI” to “CV” in the topic title, as Computer Vision is more accurate than Artificial Intelligence and it’s the term we try to use for what iNat does)

[suecar]

OK, thanks. I was in an area with relatively few moth observations, so that makes sense. I’ll try it without “seen nearby”.

[teellbee]

I may be doing something incorrectly, but when I use Identify, the suggestions are grossly bizarre. I do not have that strange behavior when I use Explore.

Identify:

[wildlandblogger]

I had this issue the other night, uploading photos without location metadata. Once I added locations manually things improved drastically.

[ellyne]

Have you tried changing the Source filter from “Observations” to “Visually Similar”? The identify tab defaults to showing you the most common species nearby in my experience and you have to change the setting to get the cv suggestions :)

[teellbee]

Thanks, I will check that. it’s already set to Nearby -edit sorry, meant visually similar- for Explore, but I can check if that’s so in my identify mode.

[tiwane]

@ellyne’s right - iNat’s showing you observations in that area, not visually similar observations.

[bernhard_hiller]

Those computer learning thingies try to extract some features from the images and then find coincidences / associations with already identified photos. Unfortunately, most systems cannot tell us about the “features” it used when it came to some conclusion.
When uploading photos of flowers, insects visiting these flowers are often suggested.
But the “features” may be rather abstract and hard to guess for a human trying to find out what went wrong.
Perhaps future algorithms might be able to explain their “thoughts” (OK, let’s keep dreaming…).

[DianaStuder]

but for teellbee’s camellia?

How is it useful for iNat to suggest - mallard is most commonly observed here?

Better for iNat to say - dunno what that is, try Visually Similar instead.

[tiwane]

It’s supposed to be used in conjuction with the Taxon filter. So for example:

Source: Observations
Taxon: Genus Rosa
Place: Sacramento County, CA, US

And that can narrow things down for you.

[DianaStuder]

Thank you - that explains why it suddenly works so much better, from a rough ID (taxon filter).

[euderma]

Here is an odd one that I had recently. A suggestion of coral fungus for a picture of a fly. Sorry for the dark picture. I took a picture of my computer screen with my cell phone to document it, but did not know what to do with it.

[royaltyler]

This a fairly new issue I have noticed also. It’s never done it this bad so something changed. I never had to change anything to get good suggestions. Must be from a recent update

[fffffffff]

It did it at least since spring, recent update had bugs that were already solved, but this is just a weird way system works, which imo shouldn’t work that way.

[DianaStuder]

following @tiwane 's advice to activate the taxon filter first.
Start with Insecta (or Pterygota or Diptera) … then activate CV for useful suggestions.

[tiwane]

@DianaStuder

following @tiwane 's advice to activate the taxon filter first.

That’s only for the Compare tool on an obs page or Suggestions tab on Identify

@euderma

Here is an odd one that I had recently. A suggestion of coral fungus for a picture of a fly. Sorry for the dark picture.

Did you enter location/date info before you asked for a suggestion?

@euderma

Sorry for the dark picture. I took a picture of my computer screen with my cell phone to document it, but did not know what to do with it.

Here’s a page that has instructions for how to take a screenshot with your computer.

[raphetnature]

Here stange results I had yesterday for a beetle photo.

Here the results I had today for the same photo.

For some reason, the results of the CV were strange yesteday, and good today.

[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.

But lately, i have had some weird ones that dont have that excuse on the upload page on web. I didnt screen shot as I didnt think to. 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.

Ive noticed this for the past month or two as being a worse issue than i recall from before.