Has anyone else noticed that the system generated id suggestions seem to have a problem in the last week or so? I use the suggestions to get me somewhere near / closer to the id for observations I can’t id from memory or logic. Recently the system’s top suggestion for a little insect I found was wedgetail eagle. Yesterday I loaded a dirt-colored seed bug and got bar-shouldered dove, straw-necked ibis, noisy miner, three-lned snout moth, lantana beetle and false cinch bug in that order. This morning, I loaded a clear insect on a white sheet background and got a range of plant options. I can’t see any obvious correlations between these suggested ids and my photo. Any one else experiencing this?
Are you using Seen Nearby (if nothing matches CV will still bring ‘wrong’ suggestions to you)
If you click Visually Similar - you need to be cautious about Distribution (looks good, but comes from Elsewhere) Try going up the taxon levels till Distribution agrees.
If you have something CV can’t match, it is good to start with the broad ID you know.
Which butterfly / caterpillar? Start with Lepidoptera
then try Seen Nearby
or Visually Similar
CV is a good tool adding species each month. But it is a tool, and we have to use it thoughtfully.
Once you are at a species taxon page - you can click About to see if it is already Included, or Pending and waiting for 100 photos (your obs and IDs can help tip it into next month’s new species)
I agree. Something seems different from where I am sitting with nothing different in my process. many more AI suggestions of mammals for trees and the like in the last week or two.
I have noticed recently that some NZ spider IDs via the ML have actually been reasonably accurate for once. So clearly something’s gone wrong
CV suggestions ‘always’ offer animals over plants.
Observer asks for palms (the trees) and gets offered a palmchat (bird) first. And some fall for it. Then we have Kingdom Disagreement, and extra IDs needed.
@dianastuder even though I agree with the general trend, current suggestions are weird, every photo of mine got plant as a suggestion, there’s a conflict between what is seen nearby and what the object is, but it really can’t only suggest spiders for a spider?
I also observed a very different quality of those suggestions in the last week or so. Most prominent was a fly that really clearly was a fly in my book, but got suggested a range of different plants on upload
Then again there ar a lot of cases where I am amazed by how accurate the CV is in the same timeframe, so I just shrugged it of.
I hope that such a bug is not back:
I use c.v. suggestions for sorting observations without ID (description here).
If this feature is not reliable, it will be a mess.
Hmmm…I just offer the CV my photo. CV and I don’t discuss words with two meanings. Are you and I even talking about the same system? I am referring to the suggestions I am offered in the Android version.
I saw one yesterday - from your tree project - which is probably a small herbaceous.
But only one photo of leaves - so it is difficult for humans and CV to decide.
@peakaytea if you type
pal (first 3 letters is the quickest way to get a taxon)
do you not get palmchat first, followed by Araceae for palm trees?
(And having just tried again, this time, I do get the palms first)
This is not related to CV. This is autocompletion based on the list of taxa (a list of texts, not a tree of taxa). It is only about text processing, a very general feature not specific to iNat.
Autocomplete within iNat is unforgiving of tiny typos. One letter wrong in a string of Genus species subspecies and iNat refuses to find it. Google says - did you mean … - then I can come back to iNat and try again. Especially for missing species (not missing, just an unforgiven typo)
Gives me two ways to get to an ID, CV for the image and autocomplete for text.
I am trying to explain to you that nothing is typed
Problem seems to have gone - no problems in the last couple of days
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.