I have yet to see a true pattern for these populations, but variable insect species will populate an ID suggestion that is highly inaccurate. Once the observation is posted or an attempt to post the observation is made at a different time or date (these photo examples are one day apart), a more accurate list of suggestions is available.
If I decide to do my own research and make the best attempt of ID, I will post it and check the comment option to suggest an ID where more accurate suggestions are available.
In this insects case, I did not post the observation, created the observation the next day, and correct suggestions were available. However, one photo still suggests Yellow Fin Tuna.
Edit to Add:
I reposted an observation of the same insect, and now the observation posted today also gives the suggestions for ducks, turtles, birds, fish, and insects.
the computer visions suggestions should be based on only the first photo in any observation. it looks like your photo in your first and last screenshot shows the second photo in each case. so without knowing whether the first photo in each case was the same, it’s hard to say for sure whether the different suggestions would be unexpected or not.
to be fair, in each of these cases, your suggestions are coming with a “we’re not confident” note, which should tell you that you shouldn’t take too much stock in the suggestions provided. although i’m not criticizing your photo taking skills at all, i do think it’s worth noting that you will improve your chances of getting a good suggestion with clearer photos. in these cases, i can sort of see how a blurry insect in the last screenshot looks like a wood duck taking off from the water. and based on the first photo in https://www.inaturalist.org/observations/288622408, i can also sort of see how a blurry insect could look like a partially sliced up tuna being sold at a fish market.
…
also, it may be worth noting that if you upload with a broad identification initially, then go back and look at what the computer vision suggests on that observation, it should return only results within the context of the “iconic taxon” of the observation’s taxon. in your case, you identified this as a beetle. so any suggestions you get on the existing observation now should come from insects only. also for what it’s worth, your insect struck me initially as true bug (possibly some sort of nymphal form), not a beetle, but i don’t know anyhting.
This isn’t true on the mobile app – the suggestions are for whichever image is the one you’re currently looking at, which is why they reload and change when you switch between images while adding an ID.
These computer suggestions definitely seem abnormal to me – I’ve added very similar-in-quality photos on the Android app before and never got ID suggestions as wild as “yellowfin tuna” or “wood duck” on a photo like this. I’d expect “no suggestions” before a suggestion like that. It does seem to me like something is glitching out here.
ok. you’re right here, and it looks like you can also filter your suggestions during upload.
since that’s the case, then it does seem more odd to me that the first and last screenshots seem to show the same photo but different suggestions.
it’s possible that a location difference could cause different suggestions, as tiwane is suggesting, but i think i would investigate differences in the image actually being sent to computer vision to be evaluated. i’m not at my computer to investigate more easily, but i would guess that maybe if you’re cropping the photo as part of the upload process, or maybe if you rotate your phone or pinch zoom or trigger some other condition that makes the Android app confused about how it crops and resizes the image before sending it to the CV, or maybe somehow the original image is a given size but the actual image you add to the observation ends up being a smaller version (is that even possible?), then that could theoretically be a source of differences.
your app logs might reveal if there were any differences in the images sent to the CV between your first and last screenshot maybe.
