Yes, i was just referring to the case where one photo doesn’t show the organism (but at least one other type of evidence, either photo or sound, does). So it wouldn’t be a totally invalid observation.
I guess explaining policy is the best approach, but man that’s a pain if there’s a lot (hopefully there aren’t!)
Oy. OK, if it’s even at that level of photo unacceptability, and with other feedback here taken into consideration, would this then be a true guidance statement?
“If your observation is in audio format, do not include any photos that do not clearly show the organism making the sound. If your subject is not visible but you wish to show, for example, the tree from which a hidden, recorded bird is singing, you can make a separate observation for the tree and link the two observations together by [insert best guidance about using observation fields or whatever]."
You say it “should include the organism” but we can now tag an animal observation (under Evidence of Presence) as scat, track, moult, bone, feather… so that’s not strictly true.
Although I realise at this point there is not an option to tag as ‘habitat’ (and since the tags apply to the whole observation, not individual photos, such an option probably wouldn’t work that well anyway.
To clarify, later in that thread kueda does indicate that it’s his personal opinion. Clearly a strong opinion and worth considering given the source and reasoning, but not official staff guidance at this point.
We have several statements here that a habitat photo without the organism is unacceptable. This seems to be based solely on the possible detrimental impact on the AI function, which could conceivably be confused, although I’m not convinced one photo would cause a problem.
We also have several statements that people find the habitat photos very helpful. I’m in that camp.
It’s my understanding that the AI only uses the top photo of an observation. Wouldn’t it be possible to include the habitat photo as “behind” the audio recording so it wouldn’t be used by the AI?
Then iNat has to have their own programm to show these, how people are meant to id things like Locustella or empids? In many cases it’s not what you can even hear by ear. People don’t know you can add photos to comments, etc., so it’s solely on iNat that it’s happening.
To the main topic I added them before for projects that required photos, now I deleted most of those, but not all as it takes time to transfer photos, I’d recommend adding all aditional info into description and comments. People are adding meaningless photos to hide gore or even photos that depict temperature, much less informing about species, but really all should’t be added to the observation itself.
@janetwright cv learns from all photos chosen for the process (it add ids based on the first one), but for taxa with many obs it shouldn’t be a big problem.
For anyone wondering, you can embed an image in an observation description or comment using this code:
<img src="image link here" >
(an image link will end in .jpg, .png, .gif, etc.)
If you want the image to be clickable to go to the host website or a higher resolution version, you can do:
<a href="website link"> <img src="image link" > </a>
It is definitely a few extra steps and requires you to choose a reliable external website to host the photos, but once you have the process down it’s not too bad. It can be useful for ID explanations in comments and journal posts as well.
Thanks, I usually do specify that. I clarified my comment above.
For the most part I think a note provides clearer information about habitat than most photos. “in Salix sp.”, “Arctic tundra” or “Miombo forest” speak volumes and don’t take a lot to enter.
Anyway, the constraints on iNat are what they are and for a free site with no advertising and operations dependent on a small army of volunteers it’s pretty awesome. Not being able to post habitat shots because of it screwing up the CV is a pretty reasonable constraint with pretty limited implications for the site’s overall functionality.
I do like the idea of automatically generated spectrograms.
Yes. They do not contain “evidence of the organism.”
All of which contain the DNA of the organism, hence are direct evidence of the organism.
According to the field guide, empids might not be identifiable on migration anyway, since they neither sing nor stay in their typical habitat. But the same field guide does include (subtle) field marks to look for if you can.
Well, they don’t have to be on migration? Anyway it’s just an example of taxa where sounds are important and spectograms are needed (esp. if you want another id quick and not after 3 years). And even if it’s not needed for id it’s useful to just see it and for other things like finding out which species bird is mimicing and why, e.g. some warblers use a lot of African birds’ signals that they hear when wintering there and other species only mimic European species they hear at summer.
I tried doing this but then found that iNat forces the first available photo in front of the audiorecording. This is a problem.
All of which contain the DNA of the organism, hence are direct evidence of the organism.
While bone, feather, and moult do, scat might not, and tracks do not. An audio file is functionally similar to a track, just that it’s not visual evidence but is auditory instead.
Habitat isn’t an evidence of the organism itself, while tracks and voice are (if we talk about fair game).
Habitat is not evidence of the presence of an organism, but it can be evidence for an identification. If two species are essentially identical except for microscopic features (or even just DNA), but they occupy distinctly different habitats, then a habitat photograph would be extremely useful to have in the sighting, both to identify the specimen and to help other people identify theirs (especially with people like me who don’t know the first thing about plants or habitat types!). I fully understand not having it as the first image so as not to mess up the AI, but we shouldn’t discourage the inclusion of habitat photos entirely. And regardless, is there really a difference to the AI whether a photograph contains a single pixel of an unidentifiable far-off organism (evidence of organism in the photo = yes) or no organism at all (evidence of organism in the photo = no)?
I’d think the cases where someone would record audio of an individual and at the same time observe track, scat or feather at the same site that is unambiguously due to that same individual would be too rare/edge for it to be worth qualifying the general statement that way.
Someone could more commonly observe a nest or home of the individual for the record, so that might be worth a qualification? But that particular evidence doesn’t even rate “Evidence of Presence” level (yet)…
This possibility is mainly why I was planning to “up my game” by adding photo to audio. :)
I’m wondering if a work-around might be to upload a blank picture (e.g. white square) as the first one, followed by a habitat shot. Would that still confuse the AI? Not sure how that would show up on the Identify page though. Would it still indicate that there’s a sound file for ID?
The computer vision uses all photos in an observation for training. It’s only for providing an ID that it only uses the first image.
Also, all images in an observation show up in the image gallery for a species.
I’m still having trouble with this. For example, with many parasites, information about the host is needed to ID the parasite.
For many gallformers for instance, I need to include pics that also allow (some level of) identification of the host tree. Good pics with characteristics of the host tree won’t include the gall itself (form of the tree, bark, etc.).
I guess one workaround is to make a separate observation for the host organism as well and link them. This is more work, but doable. But if this is official iNat policy then it needs to be made more explicit that that would be the only approved way to show hosts for parasites. Or at least I wasn’t aware of it before and lots of parasite observations are currently made this way (with host pics in the observation and not as a separate observation).
I also just kind of disagree with this approach at a philosophical level: the habitat an organism uses can be a character/trait of the organism in a very meaningful way. And as others have noted, in some cases as in habitat specialists, the habitat can be the key feature that distinguishes one species from another.
This is true, but I’m not sure that it matters functionally. If the focal organism is present in a picture but tiny/barely noticeable (which is allowed) vs. not actually present (which isn’t), it’s not going to matter for AI training or species galleries.
The AI will be trained on the majority of the photo (in most cases, assuming it can’t pick out itty-bitty thing that is the focal species). In fact, the AI should be capable of using habitat info in it’s neural net (or whatever it is) just like any other type of info.
Similarly, with the photo gallery, an image with the focal organism very tiny/where you can’t tell it’s the focus will still be confusing to users as they won’t immediately be able to pick it out. So I don’t think the distinction between “no organism at all” and “small, unidentifiable organism in its habitat” functionally matters much.
I would be very happy if there was some acceptable (and easy) way to include a habitat shot in normal observations (ones that are not audio files) as with mollusks, (which can be freshwater, terrestrial, estuarine or marine), the kind of habitat is extremely relevant in ID-ing.