Has there been any more work on allowing the same observation to have multiple interacting species that can each be IDed by the iNaturalist community and the observer can choose from a list of types of interaction? That would be amazing! Thanks for all the great work you do.
Long time lurker, first-time poster (been an iNaturalist member since 2012). I am working on some smart camera machine-learning enabled classifiers on ad hoc LoRaWAN networks for remote field deployment, for a USDA Conservation Innovation Grant project, which is open source.
Boy, it sure seems like you would be the right person to talk to about making our back-end database play nicely with existing major knowledge engineering systems.
We poached a senior engineer from Microsoft, have professors from three universities, and an excellent backend database hacker from a Medical school. We want our project to play well with iNaturalist data structures. Are you available to help us understand how we can be most compatible?
You do not use sounds (animals, bats, grasshoppers, mouse(birds).
I can not find the link but an open sound system (there are several) is building one up from bat sound detectors (including backround orthoptera).
There are open systems for it.
and also one for insects but all use theire own specific recognition to prevent unesccary species which do not occur on these sstems.
just for reference, i’m thinking this is the grant project: https://vetmed.tamu.edu/news/press-releases/texas-am-research-team-receives-usda-grant-to-develop-innovative-wildlife-monitoring-system/.
i sort of wonder why they want to “play well with iNaturalist data structures”? iNaturalist isn’t a data aggregator, but something like GBIF is. so it might be more important to understand how to play well with GBIF data structures, if they’re intending to send that data to an aggregator. i suppose it’s possible that the intent is to send data to iNaturalist so that they can leverage the iNaturalist computer vision (and maybe community) for species classification, but that seems like a bad workflow. so then if they needed a computer vision species classification model, then maybe it would be better to train their own. and if they wanted an iNaturalist dataset for computer vision training, the AWS open data set would be the place to look for information, i think.
Wow, this has been around for a while. I think it would be simple enough to just draw that data from filtering certain fields (e.g. Host, Host Plant, Feeding On). In my thoughts it would be more important to keep it to unique interactions, e.g. an insect on an exclusive host, rather than a bird eating any random insect. But that it is hard to filter without numerous datapoints.