I have written code to submit images from iSpot observations to the two APIs. For FASTCAT-Cloud (which identifies UK mammals) only the primary image is submitted but for Pl@ntNet, their API can look at more images, so there I send the primary image and up to four additional images. You can immediately see that its important to present iSpot with the best image of your observation first - and that needs to be done at the image upload stage of the submission.
With FASTCAT, the images are submitted first and the identifications are requested 24 hours later; with Pl@ntNet, identification is done on the fly. In both cases, any existing identifications are not considered - just the âgroupâ tag (Mammal or Plant).
In both cases, the suggestions that are returned are parsed and the code builds a comment that contains up to three of the âbestâ (highest ranked/most likely) suggestions, each consisting of a thumbnail image and the scientific and common names, which are linked to our own species dictionary entry. Sometimes there is no match to our dictionary - this is fairly common with Pl@ntNet because their system contains images of non-UK-native species.
In neither case does the code add an identification to the observation - just a comment - so the users are free to consider or ignore the suggestions as they see fit. It may be that none of the suggestions are correct - remember these systems rely entirely on image processing and do not consider the text description or the location of the find.
At the moment, we donât offer any means of feedback on the accuracy of suggestions, but thatâs something we might consider later.
Just an observation: ID keys in books only work when the author of the key had the species one is looking at in mind when making the key. Beginners may not always realise that. So âwrongâ IDs are always possible whether from books or AI. âAlways consider you might be wrongâ is applicable most of the time.
Thanks for your question @pcdtreeman ⊠this iForum LIVE space will be kept open after the live session, so you can access this after. Please revisit the resources that we have dropped in here and you are welcome to read through the discussions and ask any further questions you may have
Hereâs an interesting one from the FASTCAT-Cloud system. Despite the animal occupying only a very small area of the frame, it has homed in on it - but hasnât quite got it right:
I think it is particularly the case with AI as people often think these systems are somehow âmagicâ, the plantnet system is global so tries to identify any plant from anywhere. whereas conventional books are nearly always confined to a specific region. Of course it is still the case with books e.g. fungi books for a country that claim to be comprehensive when they only contain 1/3 of the total species.
Interesting observation that IDing is a hard, even with expert keys. Gives us (who work with AI tools) hope that we can contribute something useful, and if only itâs to provide a range of possibilities to consider.
Great point @JoC, there is a similar closed world assumption, and even statistical models are primed for what is plausible and likely. But with changing climates we are more likely to encounter previously unexpected flora and faunaâŠ
So far Cos4Cloud partners have been involving different individuals and groups in the codesign and now testing of the services through a range of events and activities.
For us this is very important. The Open University is a Cos4Cloud partner and iSpot is one of the Cos4Cloud citizen observatories integrating and testing these services as part of our contribution to research, citizen science, teaching and learning.
We need your input testing both FASTCAT-Cloud and the PlantNet-API in iSpot. Currently the AI is working in one direction in that it pulls an observation and suggests options in the comments section. We need human computer interaction i.e.you interacting with both services - such as adding comments and responding to the AI suggestions: were the options useful or correct? This will help to give us feedback on how they are performing.
We would like you to do this as much as possible over the coming weeks and in September two other iForum LIVE! sessions will be scheduled, one on each service, to facilitate scheduled group discussions about testing. We will inform you about the dates soon.
In the meantime we invite you to continue to add your comments and contributions anytime.
Thank you so much to everyone for your contributions in the live session today. This space continues to remain open and we would love all our iSpotters to join in â please do read through everything and drop any questions you may have here. Everyone is welcome! We will endeavour to answer all your questions
Thanks for your time. I was led to believe that ISpot is for less informed members of the citizen science community, compared to other âreporting appsâ which I am not sure if I am allowed to mention here!. I see that it is far from that and is used by serious scientists. My interest is in identifying but also for providing records eg NBN Atlas. How does an ISpot recording get verified and does it end up on records to be able to be used eg by local environmental record centres?
I have used plantnet around the world and found it very useful but always cross reference to other sources and come to it with previous plant knowledge so I have a good idea if it is giving me good suggestions often
