Thank you! Did you upload these observations to iNaturalist? If so, can you share the observations? I can’t tell where these observations were made from your screenshots or examine the photos closely enough to understand what the model is reporting.
2 Likes
2 Likes
Thanks!
Try clicking the ‘Include suggestions not seen nearby’. If there aren’t many observations in your area, the list of suggestions could be quite off as in your case.
I did a fairly crude experiment with the Genus Eristalis about a year ago - a genus of hoverflies that is generally ‘hard but rarely impossible’ - but with a large number of obs to train on. I was using things already ID’d to genus exactly and tested the CV IDs unfiltered for taxon, and not limited to seen nearby etc, I was pleasantly surprised to see it getting the species right around 75% of the time (out of 100 obs). I just repeated the exercise (but with only 50 obs - ignoring ones I don’t think were identifiable). There were 12 errors and 38 correct IDs. The correct IDs covered 12 species on 5 continents. The errors were tyically poor photos - otherwise the correct ID was usually second.
3 Likes
That’s pretty good!
1 Like