Flagger | Content Author | Content | Reason | Flag Created | Resolved by | Resolution |
---|---|---|---|---|---|---|
bouteloua | Genus Chironomus |
overconfident computer vision suggestion? |
Oct. 25, 2019 21:38:03 +0000 | bouteloua |
see link to help with ID |
@valentyna_and_midgedoctor says: "First suggestion coming up on many Culicomorphan males, majority of chironomids and many Chaoborids are getting "id'ed" as Chironomus. Manual autorisation of the ID's might be in order"
cc @edanko @vkord @treegrow @zoology123
Link to ID Chironomus: https://www.inaturalist.org/observations/identify?quality_grade=needs_id%2Cresearch&taxon_id=129409
Hi everyone!
So my point here - Chironomus is used as a first suggestion by the machine vision algorythm for almost all Culicomorphan males with bushy antennae. I have to correct quite a lot of the "Chironomus" id, which people press first thing, all the time. So Chironomids have 300+ genera, but the majority of the images uploaded to INat are getting tagged "Chironomus", taking into the watery grave with themself any poetnetial usefulness of the observations for attempts on monitoring of freshwater insects (I know, I tried to use and clean the data). Additionaly, records of the Ceratopogonidae and Chaoboridae are getting contaminated by the false -id's as Chironomus. So the question is - is there any curatorial mechanism to tone down this overeager machine vision algorythm, or can I retratin it, by feeding it some obscene amount of the actuall Chironomus spp images?
Best wishes
Viktor
" is there any curatorial mechanism to tone down this overeager machine vision algorythm,"
Unfortunately no - the community has asked, and the staff will not change or remove suggestions from computer vision.
The computer vision model is also only trained once or twice a year. Giving it lots of good data—correctly identified observations, many photos of many different species—is really the only thing we can do from the user side.
ok, so I am just going back to correcting all the chironomid records. Thanks for the explanation @bouteloua!
@valentyna_and_midgedoctor Shouldn't you be marking observations that can't be IDed further as (No, it's as good as it can be) making some obs that are family, tribe, subfamily research grade so the AI can use that to train?
Hey @zoology123 ! Great idea, I have seen you done it for quite a few observations, but I dont know how to mark it
fyi @valentyna_and_midgedoctor we've been adding some problematic taxa to a list here:
https://forum.inaturalist.org/t/computer-vision-clean-up-wiki/7281