I’m not sure if it’s a common occurrence elsewhere in the world, but here in Australia we get regular influxes of university students posting observations for a subject with a little label in one or all of the photos. Here’s a good example where the last image has a card that presumably specifies the subject. I’ve never had any issue with these before and sometimes they can actually be quite helpful in determining a rough size for the organism posted.
However, today I came across a handful of observations where the first image in the observation has only this label and then the actual organism is only in the latter photos. I’m not entirely sure what to do in these situations. It’s clear that the observer doesn’t mean any harm but in my view this situation is not ideal, because we now have images linked to a taxon that don’t show anything to do with that taxon at all. As I’m sure has been discussed before, this could potentially be problematic for AI training, external aggregators, etc. even though it’s only a small number of sightings.
Does anyone have any suggestions as to what to do in these situations? In my experience these students typically only use iNat for the purposes of the subject and then never look at it again, so an explanatory comment asking them to remove the problematic images is unlikely to get a response. If I were to be strictly logical with it, the only organism shown in the first image is a human so I should ID at the highest taxonomic level shared by both, but this also seems like it would be counterproductive.
as an aside, the purpose of these is to ‘prove’ that the student took the photos themselves and didn’t just swipe them from somewhere online. Can be faked of course by someone determined, but probably would involve more effort/time than just finding the bugs themselves.
Perhaps just leave a simple request that they re-order their photos so that the first image shows the subject might work in some cases? Maybe a few of them will do it which reduces the problem, maybe future students will see it and adopt better habits, either way, it might not cure the issue, but it is unlikely to do any harm.
I’m sure AI can cope with isolating the subject from all manner of backgrounds so I would have thought that a bit of card would be no different to twigs and branches etc… there are so many poor quality images where the subject is barely visible so it must have some way of just ignoring them and moving on?
People often make this assumption about having rogue images in the pile.
I doubt one or two anomalies in the dataset will have much impact unless they are species with limited training data - 50 obs / 100 images or so total. Even then, I would imagine extreme outliers such as this will be pruned.
If it’s more like 500 obs/1000 images it will be so negligible. Also in the dataset might be habitat shots, other species which are part of the image… Policing a perfectly clean data-set within iNaturalist is not realistic. But the CV works pretty well nevertheless for the most part.
How so? Similarly, any manual use of the dataset would have to clean it for use in some way.
Leaving a friendly note explaining why this might be problematic and how to reorder the images could help. If this is quite obviously related to a class project and it’s possible to identify the teacher (who presumably would be the project owner), it might be worth contacting them as well. They could incorporate “don’t post the label as the first image” into their instructions to the students.
I would think the best way to solve the issue is to ensure that the educators making these assignments know to tell the students “make sure the organism is in all the photos” and for them to know they need to review all the student observations and speak with students not following the guidelines about fixing things. I’ve heard horror stories of educators getting defensive when reached out to about these sorts of issues, but educator education is the only way these “student project” observations will improve in quality.
On a side note, I think there’s a bit of a bias in how we view “student” observations, because if they’re doing everything right, they blend into all the other observations and don’t get noticed, while a handful of poorly moderated projects catch our attention when we see the influx of questionable observations. I use iNat in some of the classes I teach, and I have some professor friends who assign iNat projects as well. But we’re obsessive iNat users ourselves, so we know the system and help the students post things in a way that doesn’t jump out as “student work”. If I saw a posting by one of my students with a photo of just a card included, I’d show the student how to improve that observation when I saw them in class the next day. But if the teacher doesn’t know any better, there’s no hope of getting the students to change their practices.