There was some intervention at some point to try and resolve the huge number of misidentified S.carnaria. After this it seemed to not come up so much in autosuggest.
But now… its back ( in UK at least ) and its another we have to regularly bounce back to genus.
I don´t really understand what´s caused the shift. There can´t have been greater than 50 photos of it even when the last model was updated, so it shouldn´t be in the CV training data, yet it´s being autosuggested ?
The obs are all just in the last month basically.
I thought it might be connected to the recent change where autosuggest is only offering nearby species. But surely the nearby species are only species which are also present as training data in the CV model…?
I can’t find any evidence that this taxon was not included in the latest computer vision model. There are >900 observations with an ID of S. carnaria since the last model was released, and spot checking a few almost all of them have computer vision IDs: https://www.inaturalist.org/observations?d1=2020-03-19&d2=2021-04-30&ident_taxon_id=124550&place_id=any (and I excluded May 2021 here)
More likely the people who have typically helped with correcting these just haven’t fully kept up with the inflow. There’s a flag here if people want to use that as a space to discuss identification of this group of flies: https://www.inaturalist.org/flags/484370
The link you posted to isn´t only S.carnaria …or am I missing something?
Surely the CV model isn´t based on all identifications regardless of community/observation taxon?
When I try and search only those with observation taxon S.carnaria and search within those same dates I get 2 observations. If I search for all time from 2001 until April 2021, I get 9 observations :
s. carnaria seems to be still suggested by the current model.
For example, this observation will suggest it if you include “species not seen nearby.”
I agree with Bouteloua that perhaps people haven’t kept up with bumping the new s. carnaria observations to a higher level.
One possible reason why you noticed an improvement was perhaps connected to when iNat switched from having the default CV suggestions be only those that were seen nearby. Users can still toggle that off though and get the CV suggestions not seen nearby.
I do have a open feature request to label species that are included in the current CV model (to remove all doubt about situations like these), but it hasn’t gained that much traction.
Ah right… I am one of the people who bumps them up in UK…so stuck out to me.
But maybe I am putting two and two together and making four then.
Its been out of season here until recently so that will be part of the reason.
Then I guess although I can only see 9 obs worldwide before May atm, when the last CV model was trained, it might have been trained on more observations which weren’t yet knocked back and have been since. Hence the other part of the confusion.
Still feel the nearby autosuggest might be exaggerating it somehow but maybe this is just my imagination…
Anyone know when was the last CV model trained?
And when the shift into nearby autosuggests took place?
It’s observations with at least one ID of S. carnaria.
Gotcha. Misunderstood what you meant, sorry.
In the past couple of months I’ve bumped hundreds of these back up to genus/family, but every time I check back in there are just as many that have reappeared
It would be good to know when the next model is being trained and then have a group effort to try and fix them ready for that particular time to get it out the autosuggest.
It’s rare species level autosuggests in Diptera are anything but a hindrance.
Ideally it would just never suggest beyond genus I think.
I’m not sure I see benefit to species level IDs being on the autosuggest full-stop…
ok, so the taxon page now states S.carnaria as “pending” which is great to see
but here’s one from today with an autosuggest symbol :
I guess this has come through Seek though then (?)
All of the recent computer vision IDs for S. carnaria that I noticed were from Seek users. While it would be nice to have that model also updated, now it’s at least a manageable amount of observations
Definitely, CV seems to be worlds better now for UK diptera in general… really impressed.
Hopefully they will iron out Seek issues soon then too if that was the origin.
I now hope the CV will finally stop suggesting Xanthogramma, when it is in fact a Chrysotoxum. These are so different that I wonder why the algorithm is so bad in this respect, and both have >1,000 observations (in Europe), but the algorithm was somehow ignorant towards Chrysotoxum - although it even has more observations than Xanthogramma…
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.