A short list of suggested taxa

Sometimes the list of suggested taxa is truncated. For example: https://www.inaturalist.org/observations/95957400

Have you seen this before? Is this a bug or a feature?

Could you provide a screen shot? That looks like a normal observation page to me, I might be missing something.

1 Like

The list is short because the Computer Vision is very confident in its suggestion.

Edit: See next answer for details

2 Likes

It’s a feature, because there are no other organisms in the training set that are a good match for the image. I wouldn’t say it’s because the Computer Vision is very confident. I see plenty of cases with just one or two suggestions, that are wrong, because the correct species isn’t in the training set yet.

Editing to emphasise that we shouldn’t assume that confidence and correctness are the same thing when it comes to CV suggestions (thanks @marina_gorbunova)

1 Like

Added

It can be very confident and be wrong though, even with the species in question in training set cv can see something completely different in it.

2 Likes

Adding a few details to this specific example, iNat so far records only 4 species of the genus in the State (although “seen nearby” may be more local, and I didn’t check that). Each of the species look different, and the suggested one has far more observations. I’m still yet to learn some things about CV, but I’d assume this either means:

CV knew all the species of the genus nearby, but excluded the others due to not being similar ir abundant enough, or

CV mostly or exclusively knew the 1 species “nearby,” and it matched.

The question remains whether CV still needed to train more. Either way it gave the correct answer and no wrong ones, so that’s good.

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.