I agree it’s not easy. My personal preference would be to actually remove the ability to log in to iNat in Seek, and keep it just a privacy and game-oriented app - explaining its relationship to iNat right now is really confusing and not easy. The posting to iNat part also creates more potential for bugs, like this one.
But, a decent number of observations have been made via Seek, so it’s not a trivial decision.
We’re working on it. Having an on-device model means that the model has to be compressed and that’s not easy. I can’t say much more now, unfortunately.
I think there’s a good reason for that: many observations of beavers on iNat contain photos of beaver chews, not of the actual organism. So the model has been trained on lots of those photos. Some sort of photo-level annotation functionality would be needed to sort those, then you’d have to figure out terms to use for labeling them and how to train the model on the a mix of photo types (organism, evidence of feeding, nest, etc).
Not being able to backup Seek is definitely an issue, it’s a bummer when people lose their observations. I’d like to see some form of it.
Thanks for the kind words and thoughtful response!