The role of CV in the new app

So I have just tested the new app today and I must say, I really like it so far. I like the new look, and some of the new features (especially bulk upload) just make using the app so much nicer.

However, I have one big concern, and that is the CV.
The top CV recommendation so far has always been at genus level at most which I always thought was a great choice. On the new app however, the CV suggests species, including some that are really hard if not impossible to identify from photos even for taxon experts (some fly species, Lasius ants). Additionally, the option to add an ID not suggested by the CV has been made quite obscure by just being a small unlabelled icon in the top right corner.

As an identifier this worries me because it could make observers blindly trusting the CV far more problematic than it is now and it may mean more work and frustration for identifiers.

Is there any reason for why this change was made?

26 Likes

is this when adding a new observation, looking at an existing observation, or both?

on the plus side, the computer vision scores are finally being visualized on a 5-point scale, although i’m not sure if most people will understand that’s what those dots represent or how to interpret them just by looking at them, without some targeted explanation.

15 Likes

Both.
Yeah I just spotted the 5-point scale while posting this. I didn’t notice it before. That’s definitely a change I like a lot

6 Likes

I wonder if a red to yellow to green scale would be more intuitive than five gray dots.

6 Likes

I like that idea. It could also be helpful to display a percentage number next to it. There is probably enough space, and depending on how the amount of green dots is determined, it may already be calculated anyway.

3 Likes

Yes. I’m not an iOS user, but the screenshot here worries me as well.

I think that the option to add one’s own ID should be far less difficult to find. It is important for people to see that they can add their own ID and are not required to follow the CV suggestions.

Many will probably use the CV regardless, but having an obvious option to enter something different instead at least suggests the possibility that the CV might not be right and maybe encourages people to think about what they are entering.

16 Likes

it’s probably more intuitive in most cases, but i assume a five dot scale is more universal for a user base that includes colorblind folks.

6 Likes

Shades of grey would work here.

If a label, e.g.: “Confidence”, were added, that would address some of the other concerns raised here.

We can think of two additional ways to add an ID in the new app in addition to CV suggestions and the magnifying glass search function:

  1. Dynamic Taxonomy rungs: click the info button on a suggestion from CV, which brings one to an About page with the Wikipedia info followed by the taxonomy, with each rung of the taxonomic ladder hyperlinked to its new About page. At the bottom of each rung’s About page is a large green button allowing one to “Select This Taxon”—so observers can choose a step in the ladder they are confident in. Users can get to any About pages of any taxon via the Search function, too. They are all links.
  2. Pre-selected taxon choices: Instead of exploring CV suggestions, an observer can choose among 11 icons representing plants, insects, birds, animals, fungi including lichen, arachnids, molluscs (this is how it’s spelled), mammals, reptiles, amphibians, ray-finned fishes, kelp diatoms and allies, and protozoans. This list must be scrolled by the user to see beyond the first four on our phone—and it’s not intuitive to us that there are more than these four choices or that they can be tapped to select them (they seem like they’re part of the “Add an ID” bar that leads to CV suggestions.) Additionally, when one taps the magnifying glass to search, these taxa are pre-populated (but one can’t scroll past amphibians; the list stops there; filed a bug report).
5 Likes


It does look like the suggestions are pretty different for observations where the CV is fairly confident about a species-level ID, while it’s pretty similar for ones where the top suggestion is just a genus. In these two side-by-sides of the new and old app with the same photo and same data, the Euglena suggestions look basically the same between the two apps, while the metalmark suggestions are very different. There’s no option to just leave it at genus level, and a bunch of very low-likelihood out-of-range options with “one dot” ranking are added to the suggestions to fill out the page.

I understand that a genus is no longer included as an ID suggestion for the CV once a species within the genus gets included in the CV, so I see why the genus level ID isn’t listed under “top suggestions” for the butterfly- but I also think I’ll miss the “we’re pretty sure” section of the ID suggestions.

Also I must be missing something about the dot-rating system, because “genus Euglena” appears as a 5-dot suggestion as “top ID suggestion” for the Euglena photo, but also appears as a 4-dot suggestion at the top of the “other suggestions” in the same screenshot. Clicking either option appears to ID the photo as the exact same taxon…

7 Likes

I’m guessing this is what’s meant to replace the “we’re pretty sure” box in the current app that shows higher taxa that are parents to the CV suggestions but aren’t themselves options in the CV model then?

I like that a lot, since I do find myself sometimes thinking “I think this is in family X, but the CV is suggesting genus Y which I’ve never heard of… I wonder if genus Y is in family X”.

I wonder if this will make newer users more or less likely to “back their IDs up” to a higher level if they’re uncertain about them.

1 Like

I was in an area without cell service today (which is true in a large portion of where i live) and i wasn’t able to figure out how to add a species ID if i had no cell service and the algorithm didn’t get the correct species ID. Is there a way to do so without cell service? If not it should definitely be added. I use the algorithm more and more as it gets better over time, but i still need to be able to add a manual ID sometmes. In the old app that was easy to do even if it wasn’t cached, you could override and just write in anything you needed to.

4 Likes

Maybe relevant ( but it’s about the Android app).
In Europe being online in some countries can be expensive…
https://forum.inaturalist.org/t/android-app-offline-species-list-to-add-ids-offline/2179

No, there isn’t a placeholder section. You could add an iconic taxon ID if you like, and/or add the actual ID in the observation note for use later if you like.

1 Like

The CV is no longer separately trained on the genus once a species within the genus is in the model. However, it can suggest higher level taxa even if it isn’t separately trained on them; presumably this involves some process of determining shared parent taxa of the top results of the algorithm. That’s how it works on the website and the old apps – it regularly makes suggestions for genus or higher levels (tribe, family). I believe it has been deliberately set so that the top suggestion is never more specific than genus, even if there is only a single species that is a good match. I appreciate this feature, because it encourages a certain amount of caution when choosing ID suggestions, and helps reduce the number of observations that need to be pushed back to genus.

I don’t understand the reasoning behind deciding to handle this differently in the new app.

5 Likes

I find the placeholder really important and hope it is added. Many many places in the world still don’t have cell service (and even when they do, if we travel internationally we don’t all have international plans).I really do prefer the new app over the old app but without that functionality i probably won’t be using it in areas without cell service. The CV gets better all the time but there are still lots of things it won’t be able to get 9especially the offline version), such as less common species. I know it’s probably just a development thing though since it’s not done yet.

2 Likes

I’m sure I would not realize that was what the dots mean, if I even noticed them . Some kind of annotation text label is needed, A tool tip pop-up for hover over would serve.

2 Likes

What’re the downsides of adding your placeholder as an obs note? To be clear I’m not disputing you, I just personally don’t see an obs note as a lesser option and I’m wondering what the advantage of a placeholder vs obs note is to you.

1 Like

If you’ve typed a species name or common name into the placeholder that doesn’t conflict with another taxon, it will automatically match and doesn’t require any extra effort when you upload it.

If you put it in the observation notes section it requires significantly more effort afterwards to cut and paste it into an ID. (leaving it in the notes section too would be annoying clutter to me)

I made around 60 observations this week using the new app in an area with spotty service so I decided not to enter any IDs at all and will do so once I upload them.

As they upload and I scramble to add IDs to my unlabeled obs, there is typically a flurry of people trying to “help” me ID when that help isn’t needed in this case - so some wasted time for multiple people.

11 Likes

i wonder if Next could force the user to select an iconic taxon as part of typing in a placeholder? that could disambiguate a few situations where, say, the same name applies to a plant and an animal, too. placeholder gets added as a taxon if a translation can be found. otherwise, it gets added as a comment on the iconic taxon-level identification.