Open larspett opened 1 year ago
the NIA classifier is going through a major update (new model due in May) and we will be using their taxon mapping tool to link the eBMS lists to their taxonomy. We will use your example here to test
Is there an issue about automated cropping? Otherwise I can submit a feature request issue about that. Beside blurry pictures, cropping vs no cropping seems to be the major factor behind getting differing responses from the classifier. The AMI trap software applies automated cropping, and cropping suggestions would be an important improvement for the app
we can certainly look into this when we get a chance, including https://segment-anything.com/
Should I submit an issue @DavidRoy ?
Ah, well, Agonopterix ciliella/heracliana is in the EBMS list but I guess it is not clever enough yet to realise that is the same as Agonopterix heracliana / ciliella
@JimBacon would be interesting with a temporary hack renaming the EBMS entry to "Agonopterix heracliana / ciliella" to see if the app classifier succeeds. Depending on the backbone structure, it might currently be 60% certain about something in the database which unfortunately lacks a name and then call it "unknown" despite being 60% certain. That is (could be) different from being 0% sure about anything and calling the entry "unknown"
@JimBacon can you add the necessary synonym(s)? I think the convention should be to list the complex in alphabetical order as the preferred name btw
We have discovered some species complexes that ObsIdentify solves but that end up as Unknown in ButterflyCount. Can that be because of those complexes not being listed in the uploaded taxonomy? This example becomes Unknown in ButterflyCount but a species aggregate (correct) in ObsIdentify. The confidence isn't sky high here but I've been up to about 90% with the same outcome.