The accuracy of the iNaturalist ID engine has changed in the last couple days. Normally if I uploaded a photo of a jumping spider, it would at least recognize it as a jumping spider, but that has changed. In the last two days, I just get “Spiders” with no recognition even with a face shot, which on a jumping spider is very distinctive.
Today it got even weirder. I tried uploading a photo of a Leopard frog, and it suggested an ID of leopard moths, leopard cats, a jaguar - everything but a frog. What is going on? Here is a screenshot of the odd IDs it had for the frog.
Hi, welcome to the iNat Forum! Once you start typing, computer vision is “deactivated”. The search results in your screenshot are based on a match of the text “leopard” rather than an attempt to ID based on the photo.
If you remove the text, what are the results? It should start with something like “We’re pretty sure this is…” and “Here are our top suggestions” rather than just showing a list of taxa.
are you adding the name of the organism to the filename of the photos that you upload? if so, what is the filename for your frog photo?
the system should try to parse organism names from filenames. so maybe if the filename is something like “leopard forg.jpg”, where “frog” is misspelled, the system might be seeing only “leopard”?
Just a comment separate from the overall topic, as it looked like you haven’t submitted the observation for the frog yet when I looked for it to add a supporting ID - this is Lithobates Clamitans (Green Frog) not a Leopard Frog.
Versions of this have happened to me. I have some photos I use for their GPS location data, with some of them named things like “moose ravine” etc. I upload these to more quickly peg a location for a trail camera photo (= the main photo for the observation), then I remove the GPS location pic. But the system inevitably identifies the organism as “moose” because of the GPS photo. For an otherwise smart system, this seems like not-so-smart behavior. Why would the AI not be set to exclude something like a filename rather than just focusing on the image itself.
Beyond this issue, in my experience the last year or two, the AI has become worse in a greater percentage of its IDs than what I recall during my first years on iNat. I find myself using more image searches (e.g. Google images), since what I’m getting from iNat is getting less helpful/reliable. And what seems to me to be very clearly-identifiable species from my pictures (e.g. mountain lions), generates inconclusive IDs with suggestions of domestic cats, etc. 'Not sure why this would be, and I still love iNat. But this has been something I’ve noticed during the last year or two.
