07. Een nieuwe ontwikkeling is het herkennen van foto`s via een web interface.

accommodate a flat set of classes (e.g. species) and not a hierarchy
nested classes (e.g. family, genus, species etc.) more than it has to
do with the fact that mainly species have are 'Research Quality'.

We're currently actively working on making the computer vision model
work on the full taxonomy and not just species (which includes making
use of data beyond just the 'Research Quality' subset). Attached is a

screenshot from iNat research collaborator Grant Van Horn's work
(https://www.inaturalist.org/people/gvanhorn) which shows a demo of
computer vision working across the taxonomy. Its confidence that this
drawing is a bird is 0.99, that its in the woodpecker order is 0.92,
and that its Pilleated Woodpecker is 0.75.

A lot of the work I've personally been doing these last few weeks with
taxonomy related coordination and functionality has been towards this
goal of making sure iNat's taxonomy can be actively curated but is
also well structured enough to work with this new model. So in case
anyone's wondering its not just because I'm totally OCD about taxonomy

Al enige tijd zijn een aantal android apps verkrijgbaar uit de obsidentiy familie welke voor een aantal soortgroepen de mogelijkheid bieden om de soort te herkennen vanaf 1 of meerdere foto`s. Deze apps hebben een volledige herkennings database in de app zelf waardoor herkennen geen dataverkeer kost en dus overal werkt.

Een nieuwe ontwikkeling is het herkennen van foto`s via een web interface.
Deze web interface werkt via de servers van naturalis :
https://identify.biodiversityanalysis.nl/beta/observation/demo
Voordeel hiervan is dat je niet hoeft te zoeken naar de juiste app
Nadeel is dat dit wel dataverkeer kost en maximaal 10 herkenningen per dag per ip-adres mogelijk zijn.

Deze web interface biedt wel weer mogelijkheden om op een aantal plekken in Waarneming.nl / Waarnemingen.be te integreren.
Optie 1:
Help de gebruiker om de soort te herkennen die op zijn foto staat.
Dat kan in Waarneming.nl nu op 2 manieren

  • Je hebt een waarneming met foto('s) al ingevoerd maar je wilt je determinatie even checken: In het waarneming detailscherm klik je op 'Controleer determinatie'



  • De uitkomst in beide gevallen kan zijn:
    'Zeker' (confident) : met een waarschijnlijkheid van >90% is het de soort , of juist een heel andere soort.
    'Onzeker' (uncertain) : met een waarschijnlijkheid tussen 60-90% is het de soort, of juist een heel andere soort.
    In andere gevallen krijg je geen 'advies'

    Optie 2:
    De webservice kan ook gebruikt worden om te valideren.
    Per soort kan dit aan of uitgezet worden.
    Per soort kan het minimale waarschijnlijkheidspercentage worden ingesteld (90-100 %)

    Als gebruiker merk je daar het volgende van:

    • Je waarneming (met foto) krijgt een grijs vinkje
    • Na het uploaden van een foto krijg je gelijk een melding als de waarneming is goedgekeurd
    • Na een eventuele wijziging wordt de validatie gelijk opnieuw uitgevoerd.

    https://www.inaturalist.org/blog/15450-announcing-changes-to-projects-on-inaturalist
    https://www.inaturalist.org/journal/ahospers/45530-66-relaties-tussen-organismes-aangeven-door-add-test-interactions

    test=interactions

    66. Interactions, Relaties, Verbondheid

    more details here: https://forum.inaturalist.org/t/add-interactions-to-species-pages/433/16 here are many ways. Have a look at
    https://www.inaturalist.org/search?q=interactions&source%5B%5D=projects

    Now a lot depends on your philosophy.

    For instance you can just add an interaction (one of the many fields): and name the other side of the interaction.
    But that assumes that you know the other organism, and that if you have it wrong you will fix it, and that if the name changes taxonomically, then you will fix it.

    see https://www.inaturalist.org/projects/specific-animal-plant-interactions

    My philosophy is that you put both as observations and then link them: that way the community takes care of the identifications, and the link will remain no matter what.
    If you follow my philosophy look at:
    https://www.inaturalist.org/projects/interactions-s-afr

    How can we get this higher up the “desired” list of features?
    Both the New Zealanders and southern Africans have projects dealing with this.
    Ours is visible at https://www.inaturalist.org/projects/interactions-s-afr 4

    Basically, we record only the active interaction (i.e. “a eats b”, not “b is eaten by a” - the latter just being the reciprocal of the first), although user pressure has resulted in us adding a passive field for the reciprocal observation, given that observations fields link only one way, so that these observations do not display their hosts) as:

    Visiting flowers: https://www.inaturalist.org/observations?field:Visiting%20a%20flower%20of:%20(Interaction) 6
    Eating: https://www.inaturalist.org/observations?field:Eating:%20(Interaction) 5
    Parasitizing: https://www.inaturalist.org/observations?field:Parasitizing:%20(Interaction) 1
    Attached to: https://www.inaturalist.org/observations?field:attached%20to:%20(Interaction)
    Carrying: https://www.inaturalist.org/observations?field:Carrying:%20(Interaction) 1
    Associated with: https://www.inaturalist.org/observations?field:Associated%20with:%20(Interaction)
    & the passive
    https://www.inaturalist.org/observations?field:Passive%20Partner%20to:%20(Interaction)

    Note that in each case the field value is the url of the interacting observation. Unfortunately we cannot use this is a query to summarize the interactions.
    We can ask
    “What flowers does the Cape Sugarbird Visit?” - https://www.inaturalist.org/observations?place_id=113055&subview=grid&taxon_id=13442&field:Visiting%20a%20flower%20of:%20(Interaction)= 3
    but we will only see the bird, and not the flowers, even though all the urls to the flowers are in the field - see: https://www.inaturalist.org/observation_fields/7459 2.

    In over 5 years of using this “set” of interactions, we have never had a request to add additional interactions (e.g. Eating = preys on = killing to eat - i.e. “killing for fun” has not cropped up), although it would be nice to have a hierarchical dictionary of interactions (e.g. visiting a flower > pollinating a flower (> for nectar, pollen, oil, gum)/robbing a flower/, etc

    I’m happy to leave the test=interactions thing available, I’m just not going to make it visible by default or integrate it into the UI. I don’t think we need to ice this topic, as I think the title sums up what we want pretty well. Personally, I think the Feature Requests category is a way to gauge what kinds of things people are interested in, and not necessarily specific implementation plans, so it’s valuable to me to know how many people chose to upvote this. In fact, I will spend one of my votes on it right now

    plant Lantana camara apparently “visits flowers of” 46 species of insects, rather than the other way around https://www.inaturalist.org/taxa/50333-Lantana-camara?test=interactions 13). Is it a functionality you can leave available, or are there reasons not to do so?

    We investigated this when we redesigned the taxon page in 2016 (yikes, that was a while ago). I just made it so you can see what we did by appending test=interactions to any taxon page URL, and I’ll use examples to explain why we didn’t develop this any further.

    The big problem looming over this whole feature is that observation fields are a bad way to model interactions. Since they represent a totally uncontrolled vocabulary, they’re rife with synonymous fields, so it’s hard interpret situations where, for example, there are both eats and preys on interactions, e.g. https://www.inaturalist.org/taxa/117520-Enhydra-lutris-nereis?test=interactions 28. What’s the difference? Why are both supported?

    Another problem is that using observation fields to model interactions means that one of the two taxa in the interaction is not subject to crowdsourced identification, so anyone can say that oaks eat humans and there’s nothing the community can do to correct that. As an example, here’s a butterfly that supposedly eats itself: https://www.inaturalist.org/taxa/51097-Papilio-zelicaon?test=interactions 16. It doesn’t, this is just due to an erroneously added observation field. Site curators could just delete this field, but that’s generally not how we like to perform quality control at iNat.

    On top of that, we really wanted to incorporate data from GLoBI 12, since we like them and we think it’s cool that they incorporate iNat interaction data, but mapping taxonomies and field semantics proved a hassle, and again it presents the problem of data that the iNat community can’t correct if they find errors.

    What we’d like to do is to make a new feature for interactions where an interaction is a relationship between two observations with clear and controlled semantics (to the extent that that’s possible). So instead adding an obs field that says an obs of an oak represents that oak eating a human, you would create an interaction and have to choose two observations, one of an oak and another of a human, and choose “eating” from a menu of interaction types where “eating” means “taxon A is putting all or part of taxon B inside its body for the purpose of personal metabolism” or something. Other users could then vote on whether that was the correct interaction type, and the two observations could be independently identified. We could try and pre-populate this new kind of data with observation fields, or at least make a tool that helps people review their own interaction obs fields to make new-style interactions out of them. That’s a lot more work, though, and it hasn’t really been a priority, so we haven’t gotten around to it.

    Anyway, that’s a long way of saying that I agree this would be cool, but doing it right will take considerable effo

    Publicado el septiembre 10, 2018 04:27 TARDE por ahospers ahospers

    Comentarios

    No hay comentarios todavía.

    Agregar un comentario

    Acceder o Crear una cuenta para agregar comentarios.