Can’t help with any of the logistics but this happens to me constantly. I upload pretty much exclusively on the app because I’m faster that way, and the site gives different suggestions almost every time. Not exaggerating. Usually it doesn’t matter because I just do something general or say what I think it is without the CV but it’s made me feel crazy at times. Glad I’m not going insane and somebody else has it happen.
1 Like
can you provide a couple of screenshots that show an example of the app and the website giving you different suggestions? are the suggestions actually different between app and website, or is it between upload and observation screens? like if you open up an existing observation in the app and the website right now, are you saying that it will give you different suggestions between app and website?
1 Like
I cannot do a test in another area effectively.
I’m in an area that iNat thinks is local for E. caespitosa. I can do a test from San Francisco and San Mateo counties, and there are no observations of E. caespitosa in those areas, but iNat still thinks E. caespitosa is there. The wrong id observation I saw this morning was in San Francisco.
When uploading an observation on the app, it will give me different suggestions than when I go back to that observation on the site, after it’s been uploaded. So, while uploading I’ll be given A, B and C as suggestions, then if I look at that same observation on the site, it’ll suggest something like B, D and E. It’ll have some of the same ones sometimes but usually they’ll both include different things from one another.
I’m not sure if it does that same thing when checking an observation already posted and then checking suggestions. For example, if I check an observation my neighbor made 2 months ago on the app, I don’t know if those suggestions would be different from the site’s suggestions. I just know that uploading on the app vs looking at that same observation on the site will yield two separate sets of suggestions.
If I use the same observation from the experiment:
https://www.inaturalist.org/observations/120821320
When I open it in the app and “Suggest ID” it gives me:
When I open it in the browser it gives me:
So what happens when you open this observation in the phone app and you try Suggest ID?
I tried to replicate the issue with an observation of a bird but it was a pretty clear shot and the suggestions were the same on both the site and the app. So, I used a picture my son took earlier of a bird wing (https://www.inaturalist.org/observations/120999676) and I figured the CV wouldn’t be as confident with that and it was not.
Here’s the suggestions when I was uploading it on the app:
Here’s the suggestions when I viewed the observation on the site after I finished uploading it:
1 Like
as noted before, i can’t reproduce your #6:
if you open your observation via the Explore page in the app instead of from your observations, do you get the same results?
…
what are the suggestions that your app gives you when you go back and look at the observation? here’s what i see on Android:
one interesting thing about what the Android app isn’t doing for me is that it isn’t providing just bird suggestions (since the observation taxon is a bird), which is something the website will do. so that’s definitely a different behavior. (i get the same suggestions you do from the website.)
Now it shows the same as it did on the site. Odd. So, same as the site but different from what the app initially said and also different from what your app says.
But you look at observation on website when you already added an id to it? Then that’s not odd that app gives you the same id after you uploaded it.
It’s not odd that the app shifts the CV suggestions after I assign an ID to it, but my ID wasn’t anything that was suggested. I question how it bounces around, I guess. And why it seems to bounce around accordingly to what I do but not what others do.
2 Likes
When I go back to this observation on Android and try to suggest the id it now shows me E. californica.
I created a new observation with the same picture
https://www.inaturalist.org/observations/124352173
and it suggested E. caespitosa.
I reopened the observation and tried again: still E. caespitosa.
So now I have two observations with the same picture and same first id, when I use Android to suggest an id I get different suggestions.
Using this picture.
i still can’t replicate. can you take a video of yourself creating a new observation on your phone?
that, along with sending the log files from your app to staff, will probably be helpful in this case.
Did you try to replicate by using the photo and creating a new observation or did you open up the observation added at 11:35pm PST on July 2?
On that observation I just tried moments ago and it still suggests E. caespitosa.
Do I send any info to “help@inaturalist.org”? What subject line should I use to get correct attention?
I tried only quickly, but I did not see how to collect log files from the iNat Android app. How do I do this?
both. i can’t reproduce either situation.
it’s worth noting that location on your latest observation is a bit different than your previous observations, and that creates a situation where the initial suggestions i get on the app get show E. californica first, but then if i click the “include suggestions not seen nearby” button, i get E. caespitosa first. (on the web, E. californica shows up first regardless of whether suggestions include not seen nearby or not.) so there’s a slight difference here, but it’s not the same thing as what you’re reporting.
see https://forum.inaturalist.org/t/app-crashes-in-explore/23889/3. you can also send your original version of the image in the same e-mail, and maybe that will help someone figure things out.
that said, i still strongly suggest that you capture your exact process as a video. troubleshooting this kind of thing probably isn’t exactly the highest priority thing. so if you can actually show some strangeness occurring without someone having to spend the time to try to reproduce the issue (which, again, i can’t reproduce), that’s probably going to increase the chances that a developer will actually sit down to try to troubleshoot.
Thanks. I’ll try to gather logs. Any harm in clearing the debug logs before I try to reproduce this?
By the way, the reason I was adding this scenario to this discussion is because I have noticed many observations that were initially identified by the observer as “Tufted Poppy” but really look to me and others as the “California Poppy”. Many people know the common name: “California Poppy” but I doubt many people have ever heard of the “Tufted Poppy”. I doubt many users would intentionally skip over a top suggestion of “California Poppy” and instead choose the second choice: “Tufted Poppy” so I am guessing the “Tufted Poppy” was the top choice in many or all of these cases.
With regards to “not seen nearby,” I remember several months ago there were a handful of E. caespitosa observations in Contra Costa County (most near Mt. Diablo) but I noticed there are none anymore. I assume someone took a close look at those and corrected those to E. californica. I don’t know if and how those former E. caespitosa observations would have affected the current suggestions.
i would tend to just keep everything, in case the staff developers see a reason to go back to the other observation you created today. but if you want to keep just the essentials, there probably shouldn’t be an issue, as long as you’re able to reproduce the issue when you create a new observation.
Logs sent to help@inaturalist.org on Jul 11, 2002, 2:12 pm PDT.
Not sure if it has something to do with it being Eschscholzia and the difference between E. californica and E. caespetosa being a very small feature.
Specifically, the little ring at the bottom of the pod, where petals attach. I’ve seen a few observation discussions where people were unsure about which of the two it was.
Per this Jepson key if you look at 1 and 1’ the difference between the two can be as minor as 0.2 mm, maybe iNat’s CV/AI struggles with it too?
My concern is the fact that the CV gives different recommendations for the same picture at different times and the id is platform dependent.
With the same picture, if I use the website the recommended id is always E. californica. If I use the Android app on my phone the initial recommended id is usually E. caespitosa. If I check the same observation for a few days the recommended id is still E. caespitosa. After some number of days (not sure how many) the recommended id changes to be E. californica.
I’ve created multiple observations with the same picture and I got the same behavior at least 4 times. Each time the initial recommendation is E. caespitosa. For the next day the recommended id is still E. caespitosa. If I come back to the same observation a week or so later the recommended id changes to be E. californica.
Nothing weird happens on the website. Only with Android. I don’t have an iPhone to compare.
I suspect this happens with other people, too. Keep in mind that “California Poppy” is a commonly known flower in California; it is the California state flower. I’ve lived all my life in California and until I used iNaturalist I never saw a reference to the Tufted Poppy. There is almost no way that a novice user who lives in California will see “California Poppy” as the top recommendation and “Tufted Poppy” as the second choice then pick “Tufted Poppy”. (Except maybe for some people who live in the Sierra foothills.) I suspect the novice users merely pick the top choice which means “Tufted Poppy” would be at the top. [Disclaimer: I have no proof of any of this…just my speculation.] I have seen many E. caespitosa observtions when I use the website to get the CV recommendation I often get E. californica as the top suggestion.
I tried looking at the 4 recent casual observations in both Android App and iNat website. I only got Eschscholzia californica as the top suggestion after Eschscholzia.
I did notice that metadata user comment for these two was a little different (Logged into Website - Click Photo - Click on (i)):
Not sure why it would matter to CV models, though…