I noticed that some observation identifications within Muscoidea are a complete mess. For example I have started to clean up “Musca domestica” species, where a lot of bad photos are around and around 90-95% of all observations are misidentified (not clear if Musca domestica - because of bad photo quality or lack of knowledge). It would be fine to get some help from others for the clean-up.
Michael
Hi Michael,
Yes, a lot of contributors think that every large robust fly in an urban setting is a “House Fly” and therefore Musca domestica. I’ve tried to work on the mis-IDs on Western North America when I can. Fortunately, many are not “research grade” - perhaps a focus on correcting the RG ones would be efficient?
Ken
is it possible to change that common name, since it’s a general one that seems to colloquially refer to more than one species? It might cut down on the continuation of this problem.
Check also that it isn’t a CV issue. Bad IDs tend to proliferate across the globe because of CV suggestions that then cause the Identifying tool to also offer this as an option.
@charlie we have the same problem with spiders, where certain common names seem to lead to mis-identifications as people (presumably) type in a descriptor in the ID box - “Brown House Spider” “Black House Spider” etc. - neither of these is found in North America so I sort of assume people start typing “Brown spider” and then click the one that comes up. But they are accepted common names so I’m not sure it would be proper to remove that association even if it may lead to fewer bad IDs.
This was an issue before the Computer Vision was in common use although it’s probably gotten worse since. Some of them are easy to clean up, but some (like Steatoda capensis “Brown House Spider”) are difficult because there are other similar-looking species that do look quite similar in photos so it’s harder to make informed corrections. I have a browser bookmark list of about 10 species that are not found here but are commonly suggested by the AI, and I periodically clean them up, but it doesn’t entirely stop the AI from suggesting them.
I’ve also had to clean up some of the Brown House Spider obs, which were appearing all over Eurasia. The trouble is, if one doesn’t vigilantly clean them up, ppl not only are tempted to select the similar looking wrong spider, but they see that there are other obs of that spider in the area (all mis-ID’d), so one can hardly blame them for thinking it the correct ID.
I try to subscribe to taxa that are commonly falsely identified by computer vision, like Dysdera crocata, so I can see when new ones are posted and at least add a disagreeing coarse ID and prevent the AI from being trained on those images.
Sarcophaga is another bad one. All large, striped flies seem to end up there. The limitations of iNaturalist become obvious when comparing it to a well-curated website like bugguide. I’ve gone through a few smaller insect taxa, but the chore is overwhelming, especially when 90% of photos on here are “frass” quality. As a professional entomologist, I would love to contribute more, but it’s simply not worth my effort until the ID protocol is reformed here.
Maybe, automatic suggestions should be disabled for certain taxa or limited up to family or tribe level.
Maybe there should be some community based quality check on the photos before any ID is possible,
so that only observations with good quality photos are entitled to be IDed. (this may also be limited to certain taxa)
I think I read somewhere that CV training is only done occasionally, since it’s computer-intensive? Does anyone know if it’s done across the board, or by group? If the latter, might it be worth a call for people to work on those groups a few days or weeks before hand?
AI is never going to be able to accurately identify Diptera on here. I’ve trained for years at the task, and I still can barely do it for some groups, even with a microscope and dichotomous key in front of me. Add to that the poor quality of most photos on here, and the difficulties are compounded further. The fact that a user can upload a blurry photo of a fly, have iNat suggest it as a Musca or Sarcophaga, and then have some other user without adequate expertise “agree” the observation to Research Grade is ridiculous. This is a fundamental shortcoming of the way this site works.
Yes, research grade IDs and ID down to species level should be possible only if the photos fulfill some quality requirements.
For example
*) Wing venation, bristles, head etc. visible and in focus (not on one single photo, but on the set of photos)
*) Sufficient contrast and color
*) Insect clearly recognizable that it could be identified at least to family level
Only if those quality criterions apply, it is possible to set genus or species as ID.
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.