Platform: web
App version number, if a mobile app issue:
Browser, if a website issue: Chrome
URLs (aka web addresses) of any relevant observations or pages:
https://www.inaturalist.org/observations/345475817
Screenshots of what you are seeing:
Description of problem:
Step 1: go to inat obs
Step 2: click suggest id
Step 3: see that the top option is a butterfly
The CV suggestions are limited to the iconic taxon of the observation. In this case, it was already identified within Insecta, so the suggestions will only include insects, not plants.
Would you say this is not a bug and is expected of the CV/AI to do this?
I didn’t write the code, but my understanding is that it is intentional and expected.
The Compare button works slightly differently, but if you wanted visually similar plants, you can find them with Compare.
if you think it’s a plant, just input the plant you think it is. you don’t have to select from the CV suggestions.
Yea I was raising a “bug”/ unintended behaviour for the cv/AI :)
Of course I did the necessary to identify it to species level in the order, Odonata already.
I thought it might be useful to highlight this? Idk… maybe reinforcement learning of sorts?
i don’t really even understand your complaint. it’s not that unreasonable that the CV might interpret the dark triangular thing as a butterfly. it’s not an unreasonable suggestion.
It is to me. I don’t understand your pushback either.
It also doesn’t help when the layman just trusts the cv for it to be a butterfly.
To me, this is a very severe lapse in judgement.
The subject is small in the frame, which inevitably will result in lower-quality suggestions. You can suggest to the observer to crop their images if they want better suggestions. Remember that the CV does not “know” anything. It is simply returning the identifications that have been added to other photos that are most visually similar to this one. In this case, that’s a large leaf with a small insect on it.
Imagine if every incorrect suggestion was reported as a bug. The bug reporting system would become overwhelming and meaningless. This doesn’t help the developers at all. Besides that, they could track disagreements between CV suggestions and community IDs programmatically if they want.
Sure. I don’t think I was being unreasonable when the leaf was detected as a butterfly though?
There are plenty of mis-ids by the cv that I don’t report as a bug. This one is too flagrant.
It is fine, since the community has overwhelmingly & collectively decided that the cv is already at its best performance. I think there is no need to highlight such items in the future.
People will just continue to get leaves to be reasonably identified as butterflies. shrugs
If you crop, CV is 96% certain of the sp and 100% for the Genus. TLDR - please crop first. Since the observer chose an insect, I wonder why you expect CV to ID a bit of leaf ?
This….. Should be automated on the back end imo.
But sure, I’m very sure many will disagree and say the onus is on the user not the cv/coding logic used to identify things.
What do you want to be automated ?
The cropping step. It should be seamless and automated instead of getting the user to do it.
But as mentioned, there will be alot of political red tape for such a “feature”. Which imo… is not rly a new feature… it is part of the entire feature of the cv.
I don’t want automated cropping of my pictures - most obs have various sp in a picture, and the observer decides which one they are looking at.
I’m not saying your photo will be cropped in the observation itself.
The cv should have a step which crops b4 identifying, which you have demonstrated was entirely possible.
creating a system which detects the subject and crops to it would be quite the undertaking, and would also require careful implementation to make sure it selects the right subject. to the CV this is a photo of a poorly lit dark triangle “perched’ on a leaf at a slightly awkward angle, the actual subject is tiny and out of focus. i second pisum’s sentiment that a butterfly isn’t a crazy guess here, i’m sure there are many very similar photos of actual butterflies in the database
Alright! I concede.
