Confusing algorithm suggestions

I’ve seen a number of times suggestions from the image identification algorithm that conflict. See screenshot attached for this observation, the first suggestion is a high confidence genus, the second is the only top recommendation provided by the algorithm (and seen nearby) but of another genus. How do I interpret this?

1 Like

Typically, I would look for a common ancestor between the two. My understanding is that the “Visually Similar” is a lower confidence and not as reliable because the “Seen Nearby” observations could be incorrect themselves.

2 Likes

This happens when there is a strong visual match with a species that’s not seen nearby. If you click the button at the bottom of the list that says “Include Species Not Seen Nearby”, I am sure you will see an Isagoras species as the top suggestion.

In these cases, the genus suggestion might still be a good one - I would check if there are any species of Isagoras seen in your area, which may just not have made it into the computer vision model yet. You can use that to help decide what ID to choose.

7 Likes

I’m not surprised. Algorithm suggestions are dubious for many life forms. I rarely use with moths - the suggestions are usually way off. It cannot pick up the subtle differences between, say Peridroma saucia and Noctua pronuba. I sometimes use it to get a sense of sub-family or genus, but for me (I mainly identify moths) it is not very much use.

Chris, I agree, what I find confusing is the UX which suggests that the two genera are equally strong candidates without any explanation of how the suggestions are generated. Usually I get a “high confidence” genus in the first section and a number of possible species (often in the same genus) among the top suggestions. In this case, the only two suggestions are distinct genera.

Only one Pseudophasma species is in the model (or at least qualified to be in), and none from the other genus. It can’t show you species as it doesn’t know them really.
UPD. Nope, it’s not in it yet, so none species are known by CV.

1 Like

Posting the url would help so we could test the CV, and open the obs. from the Identify pop up view and use the Suggestions (set to Visually Similar). I recently wrote a bug report explaining Suggestions give much different results from CV. Specifically, Suggestions often skip recommending genus or subgenus and instead overemphasize species. Even for groups which rarely can get species IDs, like Dialictus bees. For this reason, it would be useful if both Observation pages and obs. pages when viewed in Identify had CV and Suggestions (or to combine what both can do and add them together as “one CV” that can be accessed from either page). Also, finding and setting the filters for Identify Suggestions takes a long time and many steps (so many don’t see it), vs. CV only requires one click.

the first suggestion is a high confidence genus, the second is the only top recommendation provided by the algorithm (and seen nearby) but of another genus. How do I interpret this?

In most of the cases, the first suggested genus is more correct than the second recommendation seen in your region. But if you are not really sure which one is correct, open both in a new tab and search for the closest common taxonomic rank.

In your case, Isagoras and Pseudophasma are both members of the family Pseudophasmatidae. With Pseudophasmatidae you’re on the “safe side” (at least it’s very likely)…

2 Likes

I updated the post with a link to the observation. What does “CV” mean?

Computer Vision.

2 Likes

This is what I get in different circumstances.

On the Observation:
Click the ID box and it will show automatic CV suggestions. It appears as your screenshot shows. But when I select “include suggestions not seen nearby” the same “we’re pretty sure it’s this” taxon is displayed as the first “pretty sure” suggestion and again as the next suggestion below it.

In the pop up view of the observation seen from Identify grid mode:
Click Suggestions. In it, set taxon to order (Stick Insects) and filter by Visually Similar (location set to Panama). Here, Isagoras is the first suggestion, followed by another, and Pseudophasma is third. Possibly some suggestions are based on high abundance of Isagoras conflicting with better visual match to Pseudophasma?

1 Like

Often the most likely species has not been seen nearby. Press the button to see all species anywhere.

If I’m checking one I usually check both in CV on Observation pages, and check Suggestions in Identify-view when set to Visually Similar.

1 Like

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.