I am really fed up with seeing the Plantnet ID panels and would like them to stop now. If it is mainstreamed have I missed the evaluation of the trial including contributor comment?
Frequency of first choice not in the dictionary is considerable
Related - species offered often are not UK species
Three offers sometimes when all three are from another dictionary or two are very unlikely
Likelihood is not considered - a good feature of the app
Imagery is often poor and unhelpful
PlantNet is a useful app on my phone but this is completely different from a userâs viewpoint
This site generally is not a post -and -go place - there is often context and rationale for taxon decisions given by contributors and that is a valuable thing which the Plantnet panel potentially discourages
The review and write up has to be completed before the end of Feb, might be done sooner e.g. by end of Jan. There are large numbers of things being written at present so it is difficult to be more precise.
Several people have said they use plantnet app but donât like the way it is working on iSpot. There have been some suggestions about how it could work better on iSpot but any other suggestions may also be useful.
Personally I am always asking to save space on pages, the more compact the better so long as things are readable. In my view the suggestion from plantnet could be 1/3 the size it is now (e.g. at present it is arranged a particular way and has text about the trial), potentially it could also be modified to show more geographically relevant names perhaps with button to ask for global if it is likely to be a garden plant or if the observation is tagged with garden habitat, should not be used for species groups such as mosses and seaweeds, have an âoffâ button if you donât want to see it etc. Some of these items are possible, others may be difficult or impossible with current setup, also those are just some of my views, we want to hear views from a wider range of people.
Iâve found PlantNet useful as a starting point but I always restrict it to Western Europe - not unreasonable for UK observations. I find many of the PlantNet suggestions being posted in iSpot to be so unlikely that they become an irritating waste of space. I do recognise, however, that this area of AI / machine learning is in its infancy and that it is pretty much inevitable that we will have to go through a phase where poor results are the norm and, conversely, an accurate hit is a cause for wonder; roll on the time when this situation is reversed (and give us the wisdom to recognise when this happens.)
As a fully paid-up cynic, I do sometimes wonder if iSpot is being used as a training platform by PlantNet rather than PlantNet being used by iSpot to suggest identifications.
irritating [Thistle] is the least of it for me. @miked IF the report is already being written WHY is the trial not modified? It is apparently running blind and untended.
There is true mileage in a reformat - honest
Code it so it ONLY offers NHM Index IDs (it is not appearing Global Observations)
Code seaweeds OUT because it is seriously damaging P&Ns reputation
.
We have gone this far so why not run a SHORT (well managed) trial of a front-end filter, through which we (those interested), can go, to get a suggested ID. Newbies here would probably relish the IDea.
This surely is the aim - to have an (optional, I hope) AI-ID filter in iSpot?
Like everyone who adds a response, I LIKE the App. Iâd like to claim that I do not need it, MUCH preferring to do my own research and show it in my Observations.
The report writing people are very busy writing several large reports, the programmers are busy updating behind the scenes software (a very large task) so we have no capacity at moment to alter things.
We were previously told that a script was being scheduled manually. If that is still the case pausing the trial would reduce the workload. Even if the script is now automatically scheduled, turning it off should only take a few minutes.
We can understand that the OU may not have the resources to immediately rework the PlantNet integration; what isnât understood is why the OU doesnât recognise that the trial as implemented has failed and pause it. You make it sound as if the obstacle is bureaucracy.
In the distant past when doing such things, a few Câs in the right place in a script would soon disappear the offending call. Normal practice is surely to stop data collection when you are in the analysis and report writing phase. Thatâs what we used to do, but maybe things have changed. I would be writing the script to remove the PlantNet comments, but there you go.
I think that Iâm correct in saying that when Fast Feline makes a suggested âobservationâ of a birdâs ID, nothing appears on âchange trackerâ. I have found by chance that a couple of my bird photos have been correctly agreed but I had no notification of a âcommentâ appearing. That would be useful, and also would help to evaluate the success or otherwise of the fast cat.
Following Surreybirderâs recent comments here (FastFeline â I like it) I been thinking about how to use the current Automated Identification project as an opportunity for me to learn more and support posts and posters.
I have made a couple of comments already on « Plantnet suggestions » . One where I had the opportunity to see a « suggestion» (which was unlikely given its distribution) of a vetch from a place I am unlikely to visit personally.
It has also sharpened my thoughts on why a particular « Plantnet suggestion » is unlikely even if the location is possible, i.e. what feature(s) makes me fairly sure that it is not as PN suggests.
It seems that, at last, the PlantNet Contributions are sending out Changes Flags - about time and thanks
Is it the case for Mammals and Birds? @miked@Chris_Valentine
My first plant was today
and âdiscussion by the iSpot Admin Team, to help us decide what next.â Oct 2022
I am again VERY close to removing BOTH the âcommentâ and the tag. I find it very disruptive needing to respond - was that the idea?
.
These appear to be of NO value to the User nor the Developer.
It is TIME for the trial to enter Phase 2 and for a performance summary of Phase 1.
I may be the only one analysing results. It is increasingly correct but still has a MAJOR flaw of offering IDs not found in the UK.
NO-ONE is reading comments made after the intervention (apparently), nor are many people responding TO the AI suggestions.
.
It would be fair to assume that users in iSpot are not very interested in AI ID
I am on my THIRD AI Project, the two previous ones were discontinued though lack of interest and without SINGLE response from the Developers.
The third project is not much favoured https://www.ispotnature.org/communities/uk-and-ireland/view/project/865652/
FASTCAT doesnât send images back to iSpot - only species selections. The code will show the image from the âbestâ observation (at that time) for the identified species, which in turn is determined by the Likely ID tag and number of agreements.
Both FASTCAT and Pl@ntNet comments are static - that is, once added to an observation, they never change.
I am surprised to read, above, that David Howdenâs 2022 observation of a Shrew, uploaded to ispot, has been used by FastCat.
.
Ispot has this on Copyright. âAll youâre doing by putting your photos up is saying itâs OK to use them on the site. â
I read elsewhere, that FastCat is integrated with ispot. Does that mean itâs an ispot only app? .
.
Does it have other partners using it? If so, how does that fit with our copyright?
AI was one thread of a European citizen science project called Cos4Cloud (now finished) from which we received the funding thatâs paid for all of iSpotâs development over the past few years. We wanted to apply AI to citizen observations to assist with species identifications.
The two systems - FASTCAT and Pl@ntNet - have a different history. Pl@ntNet was a pre-existing system that citizens were already using, whereas FASTCATâs development started as a camera trap project.
Neither of these systems were trained with images from iSpot. I may not have been clear in my previous post but neither system âusesâ iSpot images - it merely processes them and compares what it sees with their own internal models. The big difference is that FASTCATâs models only contain bird and mammal species that are in the iSpot species dictionary (which was updated once during this period), whereas Pl@ntNet has a much larger species dictionary, which is why you sometimes see ânot in the iSpot dictionaryâ - these are usually species found more often in mainland Europe. The links you see as part of the comments containing the AI results point to entries in iSpotâs own dictionary; the images you see are from iSpot for FASTCAT identifications and from Pl@ntNet for its IDs.
We have looked at our terms of use a number of times over this time period to make sure weâre not using images beyond the rights we claim when images are uploaded.