Thanks for responding. Another one of mine with no PlantNet panel also had the AIAPP tag as does this one. I really really don’t mind not having the panels, I think the plant ones add almost nothing to the site. But the point of both obs and in particular of this sycamore obs was to see what Plantnet on the iSpot site offered, following running the first image through AIs as indicated in the description. Maybe screenshots of AIs are the confusion? I’ll look out for obs with AI screenshots (there are quite a few now) and with a PlantNet panel. Next time I’ll edit in the screenshots once the AI on this site has had a go using the photos of organism.
The Pl@net AI is unique in that it takes up to five images from the observation to do the identification, whereas the two FAST AIs only use the primary image. If there are screenshots or graphic included in the observation, I guess its possible that could throw the AI.
Thanks. Will try another by a different tack, adding the screenshots only as edits after Plantnet panel appears.
OK I’m with you now.
1 Like
Added another - but am not having any luck with positioning the location
@Chris_Valentine Thanks for commenting back to me here:
on one of Olearam’s obs.
Plantnet iSpot panel: The percentage likelihood for second and third choices is so low as to be not worth showing, I am interested since here
where I am sure it’s an Alder the options given seem very unlikely indeed in the location and time of year. A Euonymus found in C and E Europe and Asia, and two plants native to NZ, I think. What were the percentage chances of it being them that I could not see as a contributor? How can plants predominantly found in other countries come up so readily here with what presumably is a high chance of being correct, ahead of more likely (IRL) UK options? This is all so annoying. Grumpy face.
I’m afraid I don’t store that information.
That’s a question for Mike.
As soon as the Cos4cloud reports are finished then I hope the system will be modified so that it is more appropriate for UK observations. We have discussed that with the plantnet team but it is not straightforward since there are so many plants in UK from other parts of the world e.g. as garden plants or escapes.
On the whole, the Fastcat seems to be getting the birds correct. I wonder whether it is partly due to the high number of good quality bird photos that are available - and also because there are comparatively few extant bird species (around 10,000) compared with fungi (millions) and plants (don’t know but presumably a lot!)?
Agreed - I’ve been very impressed with how good the bird idents have been, even when the bird is tiny in the frame.
Half a million is a reasonable estimate for the number of embryophyte (land plant) species. Wikipedia’s list of phyla gives a total of 15,000 species for “true algae” (non-embryphyte archeoplastids), but doesn’t give any numbers for the chromalveolate algae.
In Britain it’s about 500 bird species, 5,000 plant species, and 15,000 fungi species.
2 Likes
Thanks for replies @Chris_Valentine and @miked.
I was intrigued by the instance of only one suggested ID for Olearam’s Water Lettuce and what you said, Chris, about the chances of being correct being so low - not much of a failsafe if even broad location is not taken into account, it seems to me!
It’s hard for a mere mortal to understand why the dictionary being used appears to encompass the world given that the PlantNet phone app can use W Europe as a selected default.
Remember that the fastcat system finds the animal or bird and puts a box round it so can ignore everything else in the frame whereas plants are more of an amorphus mass so is more tricky to do this kind of thing so they try to identify the whole image I think.
Hi Mag, I’m Vanessa and I recently joined the Pl@ntnet team. I’ll try to answer your question about Alder pic.
Probably the picture was taken from too far to be identified correctly (especially in winter with no leaves on). We tried submitting it in Pl@ntnet by cropping the image and zooming on the buds. Alder showed up at first rank. I don’t know whether a zooming function could be integrated in ISpot before submitting images but that could greatly help.
In Pl@ntnet you can force the referential into a determined flora (in that case western Europe). I’m not sure whether you can -or plan to do it on ISpot? @miked partially answered that question when he mentioned “We have discussed that with the Pl@ntnet team but it is not straightforward since there are so many plants in UK from other parts of the world e.g. as garden plants or escapes”. In fact having this option would avoid having NZ plants showing up among first options when you have choosen the appropriate referential. Can be problematic thus for people who have no clue whether the plant they look at is native or introduced.
Hope this helps.
Please feel free to ask any other question.
Very interesting that you mention zooming as the other AI system we use, Fastcat, does exactly that. It selects all the things it thinks are animals in the image then tries to identify all of them although I think it only gives the name of one of themback to iSpot as the current interface will only cope with one. It would be much more difficult to do this with plants as they are not discreet blobs like animals but it might still be possible to try.
Will have to think about how to deal with garden plants etc, might be able to default to western Europe except if the habitat selected is ‘gardens and parks’ (when submitting an observation the observer should select the habitat) or perhaps default to western Europe but also have a button at the end of the ID panel that asks to submit the request again but to global flora.
The other issue is Bryophytes and seaweeds both of which are classed as Plants by iSpot so are both submitted to Pl@ntnet, I see that Pl@ntnet is starting to do some Bryophytes but not many species in the database. On iSpot we do have quite a lot of both Bryophytes and seaweeds it might be possible to add those to the model even if it is just to exclude these groups from the ID, I am sure the ispot community could take a lot more pictures of both of these groups if necessary.
Thanks for comments both @vanessa_hequet and @miked. Welcome Vanessa.
Same stuff previously visited being repeated here I think:
I use the phone app and can select which flora, my default is W Europe. Still don’t see why this ‘trial’ started with world flora (but I know I’ve said this so many times before!) But it’s only one resource to be used to consider an ID.
When all three of the options generated are not in the dictionary, should that not be enough to just not bother on that occasion? If percentages are generated as they are in the phone app, why not pay attention to these likelihoods and omit the very unlikely?
I really don’t see the point at all. By all means make use of various apps as I and others do, but the large and often pointless panel for plants is a nuisance. Sorry 
Look forward to the evaluation of the trial in respect of plants, and indeed frankly to the end of this current trial. I really don’t mind at all that the AI in the trial could not ID my alder - I could and the point of such an observation in winter is to think about those lovely purple colours, not to wonder if an AI can crack it! The panel is too big and all too often pointless.
1 Like
Once the evaluation of the EU-funded project is complete, we can at the very least trim the text, but at the moment we must continue credit the organisations behind it.
By saying “Look forward to … the end of this current trial” you’ve actually contributed very usefully to the AI part of project.
It’s not all bad news - the AI got my recent wild privet correct!
I look forward to one that can do spiders!
Thanks Miked for your answer! About zooming, such a function could probably not be automatized for plants as it’s done in Fastcat in fact. Still could it be integrated as an option for the user to reframe/zoom on more distinct/clear parts of pic?
Whether to put western Europe flora default or not is a setting that really depend on the type of users you have on ISpot. The option of a button for selecting habitat can be a way of dealing with that.
For bryophytes and seaweed I know developers are thinking of introducing them, at least as a rejection class for the moment. They’ll probably use pictures of GBIF but I note that we could ask ISpot community.
Thanks for all your feedback. They are precious to help us improving the performance of the AI!