I have a question, I recently came across this observation, where the observer put down an identification that although is in the photo, isn’t taking up the majority of the image. The identifiers have disagreed with the original identification and gone with the obvious one. Is this what they should have done?
Link to example observation: [link removed by moderator]
in almost all cases, the ID added by the observer should be respected as long as the taxon is indeed in the photo, regardless of how much of the image it takes up
in this specific case, however, it seems to me that the observation is a non-serious one, so the IDs of human are fine in my opinion (especially since the grass, which was the original ID, is both a manicured lawn and unidentifiable from the photo)
I assumed as much, just wanted to be sure.
I second this, although I can’t confirm the linked observation since it has been removed.
I did want to add that if a second or possibly even third organism (of a different species) is visible in the photo I do recommend letting them know via a comment which other species is visible, and that they can duplicate the observation
there is also an option to add it to one of the observation fields within the same observation but I’m not too sure how that exactly works
This can be a huge problem that alters Computer Vision.
You can ask the observer to add a cropped photo.
Yes, it’s often best to crop but I don’t think we need to be too worried about computer vision. Computer vision is trained to identify iNat photos of a taxon, not perfectly cropped photos of a taxon.
But if the plant is almost invisible in the photo, and another plant is visible, and repeatedly, it can alter the Computer Vision.
It can, but I think that’s likely an edge case.
toblerone0902,
I will resort to human, etc., when I think the observation is in jest (which from time to time is OK with me). But for most observations I judge them on a case by case basis. If somebody IDs an animal as an Eastern Cottontail and I can’t immediately see it deep in the briars where it was shot, I may take a closer look and eventually see the telltale fur colors, texture, etc… and assume that was the best the observer could get and go ahead and ID it with no qualms IF know the area. I would and have posted such with the idea it is good to show how deep into the briars rabbits can get.
But that’s not what you are talking about. Some of these come from folks new to iNat. They get excited over one find and that leads to the addictive nature of iNat (and nature in general) and they want to rack up as many new observations as possible. Most new folks, (just like me when I got started), don’t take enough photographs perhaps because they don’t “see” those other organisms in the picture they are taking of lets say a turtle. Later they see watercress, some eelgrass, cattails and hints of this and that in the picture and it all becomes “important” to them as new observations. I think these beginnings are very important but these new observers often stick around only when they are encouraged with a comment like “Hey, maybe! If you are in this spot again, can you get several closer shots and tag me so we can get this nailed down.”
On the other hand there are of course those shots where all you can get is a group shot of waterfowl for example in the middle of a lake. Sometimes it’s best to describe the organism in question and where they are for clarity. I usually take multiple shots of such, as many as 20 or 30 and later pic the ones best for each organism to use as my lead image and crop if I have to but I won’t pretend to be perfect. :)
I agree, all in all, individual observations in photos such as you describe should be pointed out one way or the other.
There’s a whole project dedicated to these kinds of observations:
https://www.inaturalist.org/projects/ignore-the-elephant-seal
It can also lead to confusion in human brains. Looking for photos of species A, and finding photos of species B…
(oh, wait, let’s peep… ok there are 3 pixels belonging to species A in the bottom-right corner, makes sense)
Mark as Reviewed. Leave no ID ;~)
There’s no way to “mark as reviewed” (and/or hide) such photos in the Photos Search results…
(Now that’s an ID for a feature request: the ability to DQA each photo separately, in order to exclude the most confusing/blurry ones from Photos Search results…)
I wish iNat would add cropping option to upload.
I always crop, since I use Lightroom on a PC (DSLR camera).
But not everyone is familiar with editing photos.
I wonder whether it could benefit from the Computer Vision thing.
For example, “Your image is very big and will get resized, which can make IDs difficult… do you want me to reframe it around the [ insect | plant | whatever ] in the center?” (with some visual hint, so that the observer can check if the CV does not go crazy)
On the other hand, over-cropping can hinder ID.
Host plant, co-occurring species, landscape etc., even the blurry outline of an out-of-focus leaf can give important clues.
The solution is to upload both : the cropped photo and the non-cropped photo.
I always upload the cropped image and the original image, unless I cropped something personal.
An interesting note about this, because some people may not understand just how AI works in these cases. AI uses all information it can, and there is often information about a species ID in the surroundings, including other associated organisms and/or information about the habitat.
So what looks like “background” or “irrelevant other species obstructing the view” might actually be critically important ID clues.
I notice this factor operating both when the AI makes correct and incorrect ID’s. For example, if I take a photo that is clearly in a bottomland area and has a lot of moisture-loving species around, but I’m photographing a species that tends to prefer drier habitats, that just happens to be growing there, the AI is more likely to guess something outlandish that doesn’t even look like the plant in question. And it’s probably because it’s using the infromation about the background.
Another example would be when photographing insects on plants; insects have strong preferences of what types of plants they tend to be on, so the plant it is on is often key information.
Similarly sometimes I take a poor or blurry photo and the AI just gets it spot-on, and I think, there’s nowhere near enough information in the photo of the leaf or bark or buds itself, to discern. But then I realize that there are some other key indicator species in the background, and/or something very distinctive about the habitat, like a picture of exposed rocky soil or something else betraying the soil texture and/or moisture levels.
It’s fascinating because the way AI works is sort of like how our own brain works sometimes, like at least with the “associative” or “gut feeling” type reasoning.
Like when we go into a habitat we are not necessarily consciously aware of everything that we’re doing, but we often mentally filter out what sorts of species to check against based on the habitat and the other things around. Like I’ll be down in a swamp and see a sparrow, I start thinking to check against swamp sparrow, whereas when I’m in a city I’m surprised when I see anything that isn’t a house sparrow.
Not sure what the conclusions are from this, but…just be aware that the non-cropped material in the image may be more useful than you realize!
There be another observation with the same photo that records the more obvious subject. Wouldn’t ignoring the observer’s ID defeat the purpose? In an ideal universe, the observer would include a note saying the other observation exists. But there’s only so much documentation it’s realistic to expect.
But if the observer can’t be bothered - there is ‘only so much’ they in turn can expect from identifiers.