App IDs species but changes to family or genus in observation?

General
1

I wonder if anyone has come across this, and if it is a bug or by design ? I have some observations made via the app (new ios app) which when I used the AI camera, it showed a species identification, and I followed the instructions to take the photo at this point, however when the observation was shown in the app, it had reverted to genus or family ?

A couple of examples (I am afraid I cannot recall the actual species ID, but the spider was something like ending in “harvestman” type spider I think)

https://www.inaturalist.org/observations/304453430

and this one (I honestly cannot recall the name, but I think it had “brown” in it)

https://www.inaturalist.org/observations/304277335

I am just starting out and would really like to provide quality observations, so I normally upload >1 photo (unless there really is no point as the second photo would replicate the original), but I am a little frustrated by this apparent change from a species ID to a genus or family ID when the observation is made in the new ios app.

Thank you for any pointers.

2

I understand why you want to get your observation to Research Grade at the species level — it satisfies your own curiosity, and it shows that you care deeply about your contribution to the data.

That being said, you’ve touched on a topic that has sparked a lot of discussion over the years, both here on the forum and in iNaturalist comment threads.

The computer vision (CV) sometimes suggests a species, when it actually requires microscopic examination to confirm. This can lead to overconfidence in the CV suggestions, especially when it skips over more accurate IDs at the family or genus level. Often, it’s better to be “approximately right” at genus or family, than to be “precisely wrong” at species.

9 Likes
3

It’s my understanding the live camera uses a simplified model for prediction, so this may be why you are seeing a difference.

But more broadly, if you aren’t able to make an ID to a certain level based on your own expertise (ie, you aren’t sure why it would be IDed as that species instead of just the genus), you probably shouldn’t ID to the species level. It’s best to ID to the level that you are able to support based on what you know, not just select whatever the CV suggests. The CV is wrong a reasonable proportion of the time. Choosing a higher level ID that you are confident in is actually a good practice that helps reduce the number of observations IDed incorrectly.

5 Likes
4

Yes, if you’re using the AI camera the live ID suggestions are coming from a) a simplified model and b) version 2.13 of the model, not the current model available online. After you take a photo, a version of the final photo is sent to iNat’s servers to get suggestions from the full model. So it may different from what was displayed on the camera.

4 Likes
5

Does this mean that the submitted species suggestions from the app should match what the website model would suggest? I’m wondering because I’ve been noticing new observations coming from Next (and Seek) with CV IDs of a species which was removed from the website model several months ago.

6

Seek definitely uses the old model, not sure about Next, but if it doesn’t have internet access it might fall back to the old model.

1 Like
7

Hi again

I understand the non-reliance on the AI model, this is the same as the Merlin app for birds really, so, given that I am a complete novice on plants, trees, insects, arachnids etc. I would like to understand how I should make the observations ? At the moment, I use the ios app to identify the subject using AI CAMERA, when it has given what looks like a confident ID, I take the picture and upload the obs. I then take further pictures (if possible) to ensure I try to cover all angles (something I am learning from the responses to my postings !), and upload these associated with the observation. I then wait for any IDs to come in, but even then I am not really sure what I am receiving.

It sounds from what you are saying that this approach does more damage than good ?

I have access to ios app and also web site (however it is a bit of a faff to upload camera pics to website). How should I be making these observations, and what learning points do I need to grasp ?

The issue I have had in the past is matching the camera photos to the observations, hence I do it the way shown above which gets round any mismatching by uploading all associated photos at the same time I upload the observation. I hope this is making sense !

Thank you all again

1 Like
8

I usually use the iNaturalist Classic app, but I played around with the newer app to try to figure out how you can add your own IDs to your observations, instead of using the AI camera.

It looks like you have two options for changing the ID you give to your observation. Both will be done from the ‘edit observation’ screen.

First option: you can select the button labelled ‘ID with AI,’ which gives you AI suggestions. You can pick from one of these, but if you’re not sure or you want to pick another option, you can tap the magnifying glass icon in the top right and type in what you want (e.g. plants, dragonflies, fungi).

Second option: If you don’t want to see the AI suggestions before adding your own ID, then you can tap the ‘search’ button on the edit screen, instead of the ‘ID with AI’ button. This takes you straight to where you can type in your own ID, without showing the AI suggestions.

If you want to avoid the AI suggestions entirely, it looks like you can go into settings, turn on ‘advanced mode’ and change the settings under ‘After capturing or importing photos, show’ to have it take you right to the edit page, instead of giving you AI suggestions first.

You might also want to try using the standard camera option, instead of the AI camera option. This option lets you take several photos for the same observation, then when you’re ready you can proceed to assign an ID. No need to go back and take more photos afterwards if you can take them all before.

Hopefully some of this makes sense, and hopefully this is actually the best way to do this. As I said before, I always use the old app instead of the new app, so I’ve just figured this all out by playing around. Others feel free to chime in if there are better / easier ways to change your ID.

3 Likes
9

This is pure gold. I will try the suggestions - particularly the multiple photos before upload.

I have to say I have not used classic app but the new app feels well designed and sleek. Thank u to all for helping me to find my way around the features, buttons , tabs etc.

Not sure if all I have learned in 2 short weeks is in newbie faq, but it was fantastic learning to bring me up to speed.

3 Likes
10

I’m sorry to see you had to navigate the meanders of iNaturalist in your first weeks, however a big thank you for taking the time to dig into how to identify stuff the right way.

The iNaturalist site and apps make a rather poor job at onboarding new users – this, together with the pitfalls and shortcomings of ‘Computer Vision AI’, is a recurring topic across several Feature Requests (little traction gained though, scarce funding goes towards other goals). Not even the ‘community guidelines’ (a link next to a checkbox upon signing up) have a dedicated ‘Good form when identifying’ section, despite purportedly caring about ‘data quality’ in their preamble.

1 Like
11

In my experience, it is, at least for me, better to take the pictures with my ordinary built-in camera app where I have more options and settings and I am more likely to take a sharp photo at the right exposure, maybe using the macro lens with the flashlight on. Afterwards (sometimes weeks later), I select the photos I like in the gallery and share them with the iNaturalist app.

In the app, if using CV, I usually check the CV suggestions for multiple photos, not just for the first one.

6 Likes
12

As others have said, thanks for taking the time to ask advice. I think your experience here is a great representation of how new users encounter the app and a concern that many folks on the forum share about the new app default settings pushing new users to rely on and submit IDs solely based on the CV. A related concern is that it is not clear or intuitive for many users how they can add their own ID that isn’t just what the CV suggests. So, thanks for sharing your experience!

7 Likes
13

Not necessarily. If the user has no internet connection, it uses the onboard model.

1 Like
14

You’re not doing any damage, but as other commenters have mentioned, taking regular photos and then using the app to upload will probably lead to a better outcome.

The one spot where it’s easy to do “damage” is identifying other people’s observations. If the observer and 1st identifier both rely on CV/AI for identification, and that identification is wrong, it will create an incorrect ‘Research Grade’ observation. This can be confusing, and difficult to correct.

The forums are full of helpful info, kudos to you for coming and asking questions!

3 Likes
15

This is my method:

I take photos with my phone and if necessary close ups with my camera. (I cannot hold my phone still enough for macro shots, but can with my camera). I do not take the photos from within the app.

Between species, I take a photo of my feet. I ensure my camera is on the same time zone as my camera (I do travel a bit!). Then I put them in my photo management software and sort by date taken.

If I only have phone photos, I will upload the observation via the app, otherwise I upload via the web. Unless it is something pretty easy, I find it easier to sort out ids on the computer screen.

Your observations have improved in the last few weeks. Have fun with iNat!

2 Likes
16

Thank you for the vote of confidence Ruth et al. As i have said, my issue with taking the photos then matching to an ID and uploading later is fine if you only did a couple of IDs. My downfall comes if you do a lot of IDs (say on a nature walk) then trying to match and remember which leaf goes with which plant or flower ! I know you would take the pictures consecutively, but (for example) if your last photo of one species was a leaf, and your first of a new species was also a leaf, that is when i struggle - worse if the two species have similar leaves !!

If anyone has any advice other than this comes with experience, would love to hear it please :-)

17

I take a ‘junk’ photo in between. The view, some lichen on a rock? Next plant for iNat.

3 Likes
18

Now I get the photo of the feet, have to say I was a little confused over this one :-)

1 Like
19

When the conditions are right, I make a lot of observations with lots of photos each - and would like to keep up as well.
The idea is to reduce time fiddling with or swapping cameras. I take overall, detail of any inflorescence, fruit, branches, leaves, stem root than macro of details than proceed to the next one with macro (if possible) first and reverse order ending in overall.
If you don’t remember later where one series of photos ends and where the other starts, I would recommend making a photo of the back of a leaf the end of a zoom in series. In a similar fashion, an overall photo of the backlit plant could mark the end of a zoom out series.

1 Like
20

I have never taken photos within the app, I find it very difficult to see images clearly on a phone when outside. Unless it is very obvious, a common butterfly perhaps, I need to look carefully later indoors. For this reason I prefer to use a camera and sit at a PC with a big screeen.

Learning to observe in the field is very rewarding. Looking a bit more deeply at the photographs can be similarly rewarding.

3 Likes