Continuing the discussion from State of matter Life limbo:
Perhaps habitat images, when deemed appropriate to an observation, can be included as an additional image to a specific observation. Cyperus javanicus is characteristically found on the margins of water, thus the third image in observation 25368804 is of the sedge next to the water, the fourth provides the context that this is a freshwater location.
For observation 2911994 perhaps the guidance might be:
I have plenty of observations that depict habitats/landscapes with easily identifiable species. Coincidentally just earlier this week I wrote to the staff to request they remove that clause about landscapes from the guidance on evidence of an organism (though I have not yet heard back).
Marking an observation with organisms present as “no evidence of organism” is an inappropriate use of the DQA. iNaturalist is not a tool to map anything, it’s a tool to map organism occurrences.
Will write an issue about the clause from the DQA, @bouteloua, sorry about the delay.
Each photo of an observation should have the subject in it, so a wide photo showing the habitat of a mushroom would be fine as long as the mushroom is in it. I’d say that’s the best way to do a habitat shot right now.
I know we eventually want to add a way to categorize photos. Then you would be able to do things like label the photo a habitat one, or a photo of a leaf, or head, etc.
Ahh, and I didn’t mean to imply that taking more than 2 days to respond to an email was a bad thing, just that I hadn’t heard one way or the other from y’all. :)
This seems like categorizing photos would be important to the computer vision training process. My third and fourth photos are problematic for computer vision training unless labeled as being habitat images. Pending photo labeling being developed and deployed, perhaps observers could be encouraged to put the optimal photo first and computer vision could train only on the first image in a set?
Pl@antNet requires the user to identify the plant part in the image prior to upload. The complication was, back when I used the app, that one could only upload a single image.
I think habitat or context photos are a serious value added aspect of observations! I had always done this and recommended to others to do it, until it was pointed out that it’s against iNat guidelines. Now, I am a sucker for a good rule, but I am also a realist, and believe that we need to use common sense too… if the context shot adds useful information to the observation, I think it needs to be there!
I experimented with an observation that had photos not of the organism itself, so technically they should not be part of the observation. I uploaded all 3 photos, then right clicked on the “context” photos to copy the image url, and then edited and added image tags into the description. Finally, I deleted (unchecked) the two photos that should not be included in the photo evidence for the observation! Voila… hosted on iNat, present in the observation, not confusing the AI :)
Brilliant! Thanks for this! I have already edited my observation to utilize your technique. I had not realized that the description box accepts HTML. Based on the guidance above, the habitat image that still contained the target organism was retained in the observation, although for computer vision training purposes that third image is of questionable value. The habitat image that did not contain the target organism is now in an image tag in the description.
This solves another problem that I had thought of: if one were to upload multiple observations all with the same habitat shot in each observation, this would be wasteful of storage space. The above technique permits a single habitat image to be “uploaded” and then the URL can be reused in each description to pull up the same habitat shot for each observation in that habitat. More efficient use of storage.
I agree with @bouteloua .
Personally, i do a lot of spatial ecology like natural community mapping and DO post landscape shots but i am doing it of ecosystems with important and easily identifiable species (if i couldn’t identify the species i couldn’t identify the natural community). So, i just post say, a cedar swamp landscape with a bunch of cedars in it. It works fine and usually no one complains (if they do you can ignore them so long as the species is present).
Every once in a while the photo comes out blurry and you can’t identify that organism, in that case i either mark it as not needing further ID or delete the photo (who needs a blurry landscape pic anyway)
IF the species is in the image, my guess is it does more good than harm for the algorithm. I have a funny story about that but i need ot go to work…
That’s really clever! I have one doubt though - can we be sure that those unchecked photos won’t disappear in the future?
I have obs that I have done this for about 2 years ago, and they are still working ok. Of course there is no guarantee they will persist, but because they are largely to assist in the initial ID, I figure the obs won’t suffer in the remote chance iNat do a cleanup of orphaned photos. And if enough people did it as a standard practise, I am sure iNat would become aware of it and be less likely to do such a cleanup, or will see the need and build in similar functionality.
This is quite a clever workaround for habitat and one that I can see as being valuable. Curious to hear the opinion of our hosts.
I can understand the downside of having a habitat photo (or two or three) within the set of photos that have been elevated to a research grade observation through community agreement. Besides the images potentially being part of the training for “Computer Vision”, I believe the images are accessed by external organizations such as the Encyclopedia of Life.
When I first started using iNaturalist and before it had the power that it now has, I used to search the images on EOL and the Map of Life to try to get a better idea of the appearance and spatial morphology of different species. I think it would be annoying to try to search for an image of a Lemming and have to filter through habitat photos - it’s enough to have to wade through poor or uncropped images (personally guilty of supplying some but always trying to improve unless it is that rare for me bird photo that I got with a 600mm lens and still had to digitally zoom in producing a tiny image hoping there was enough diagnostic material present)
It occurred to me today that computer vision is matching image artefacts, so if there were a lot of images of the sedge in habitats, then when new photos of that sedge in habitats are encountered, the matches would still happen! Sometimes it would be elements from the habitat that help make differentiation possible (eg presence of “water”)
I concur with both quotes which is why I opted to retain the third photo in the observation - with the target specimen in the image, there is potential training value. The image that had been the fourth image, the habitat shot, did not include the target, so I used @kiwifergus’s technique to separate that out into the description. Thanks for the good words!
My thoughts in general are that we should care more about accurately recording the encounter with the organism, which can (should?) include habitat (as well as nests, footprints, etc.), and less about whether it might confuse the poor Computer Vision algorithm. Robots should work for us, not us for them!
I played around with this idea a little and a couple of things came to mind:
I decided to try @kiwifergus inclusion in the description and came up with this but my habitat photo is stored with Google: https://www.inaturalist.org/observations/26760199
I also decided to try and use a the same Google Photo for my Journal Post for displaying habitat and came up with this: https://www.inaturalist.org/journal/bobmcd/25401-spectacle-lake-provincial-park
Seems to me that the Journal Post can be a little more diverse. I’m looking forward to expanding on this: Multiple habitat photos, Aerial Photos, Video links, Maps, Routes…
(I have pulled this post a couple of times because somehow I have corrupted the link but it seems to be working now. The image must be in a shared folder. I have opened an incognito window to see if it works for others and so far it does. I will keep checking it and post separately if there are changes)
Still not working predictably 
looks good to me!
A related approach that I use. “Store” the image in a Google Blogger blog post and pull that image from the blog. The photos my Android phone takes are automatically uploaded to Google Photos, when drafting the blog post I select “Add photos from phone” and can add photos to the blog. Publish the blog, view the blog, click on the desired photo. This pops up the full size image. Right click to get the address of this higher resolution photo and use that address in an HTML img tag:
My thanks to @bobmcd, I use this technique in other situations such as pulling images from my blog into the student flora for my ethnobotany class, I would not have thought of that for habitat photos in the description without his posting of his attempts to use Google Photos. Thanks!
My bad: that image WAS being loaded across from the blog and was NOT in the forum, but now the image is on the forum server and I did not do this, the underlying Discourse software did this.
I am unaware of whether iNaturalist would behave the same way, but I doubt it, and it does not invalidate the approach. The image still winds up in the description.
Thanks. I just checked via an incognito window and it seems to be working so far. I tried getting fancy with the post and added a map to the journal via a route plotting app and it seemed to corrupt the journal somehow.
