search

NERC-CEH / irecord-app

📱iRecord mobile application
https://irecord.org.uk/app/
GNU General Public License v3.0
6 stars 7 forks source link

Do images from accepted records feed back into AI training data? #273

Open Sam-Amy opened 9 months ago

Sam-Amy commented 9 months ago

This has been asked via the iRecord inbox.

I know the app uses a generic classifier, and I've heard NIA mentioned, but is there any more intofmation on this? I see quite a lot of queries about the AI, so I would like to understand more about how it works - perhaps it might be helpful to put some information on help pages @kitenetter?

kazlauskis commented 9 months ago

I think, it is one-directional, but maybe I am wrong and a mechanism exists that feeds back into the classifier behind the waarneming API.

kitenetter commented 9 months ago

Hi @Sam-Amy my understanding is that images from all verified records on iRecord (including those submitted via the app) were shared last year with the Nature Identification API (NIA) team. I don't know a lot of detail about this, and others can say more (e.g. Alba, David). I'm not sure if we have an ongoing arrangement to provide additional images going forwards.

Agree that it would be good to have more info about this on the Help pages, maybe Alba can provide some of this?

albags commented 8 months ago

Hi all, we renewed the agreement with Naturalis this year too. Last month I sent them the taxa data and images again, including the ones from last year and any updates. @DavidRoy can give more details on the agreement itself. I agree it would be a good idea to include some information in the help pages.

sacrevert commented 1 month ago

Seems like a low-hanging fruit if this is just about adding info to a help page.

kazlauskis commented 1 month ago

Agree. If someone could draft a paragraph, then I could add it to the help pages.

sacrevert commented 1 month ago

@albags seems like you are best placed to do this? Perhaps you could post text here? Cheers

albags commented 1 month ago

I think @DavidRoy might be better suited to write a post about it. :) I was involved in the data extraction for training only, but he is more aware of the agreement with the Natrualis team.