I’ve noticed that the algorithm suggest far too accurate identification than it’s possible to do based on dorsal view for Chilopoda (Arthropoda: Myriapoda). In North America, they suggest almost always Lithobius forficatus for the order Lithobiomorpha, and Strigamia or Haplophilus subterraneus for the Geophilomorpha. But right now, I’m sure that 95% of these identification are wrong (of those not confirmed by an expert). I think it would be better for the data quality that the ‘suggestion’ algorithm stop at the order level. Is it possible to ask that? I know I can get several Chilopoda experts to back me on on that.
Take a look here.
I’ve been a bit annoyed by that as well, at a more general level, and made a feature request to enable a basic level of this.
In short, not really. You can sort of force the CV to suggest things by making an initial ID and then making a second ID, which is a really clunky andanniying way of going about it, but that has limits, and in my experience it doesn’t at all work when you get beyond the very most basic and broad of categories.
The main thing is to get enough correctly identified observations to research grade so that the CV can be trained on them, and hopefully that will result in more accurate IDs.
There is another more recent issue, which is that the change to ‘expected nearby’ which is a good idea) has superseded the older ‘seen nearby’ ID suggestion, and now appears to be returning less accurate suggestions in many areas.
The iNat staff don’t manually remove individual suggestions from the computer vision model.
The best way to prevent it is by correcting the incorrect observations so that future versions of computer vision can train on better data and improve.