Description: Using Firefox 136.0.1 (64bit), since this morning about 11am South African time, when I upload obs and click on the search button to ID, the suggestion box is a drop down list with the same recommendations for everything. It does not matter whether it is a plant, insect, fungi, leaves, flowers the suggestion list is always: Spotted eagle owl, common eagle ray, sea sparkle, …, common bark spider. The screenshot is of an African Veined White I uploaded yesterday, which yesterday had the correct suggestions, but today I get the gibberish list. Same browser, same laptop, same obs.
I cleared cookies and history and restarted the browser with no change to the outcome. I tried a Firefox private window and get the same result.
The suggestion dropdown works perfectly correct in Edge as well as the Android app. I would honestly prefer to use Firefox though.
As @pisum said, it’s probably due to something int he photos’ file names or metadata, and iNat is automatically filling in the ID based on that and not using computer vision suggestions. Without knowing what the file names of the photos are, or any keywords embedded in the photos’ data, we can’t say for sure but it’s almost certainly due to those.
the other thread i pointed to notes 2 possible things (neither of which is related to names or metadata):
Firefox’s HTML canvas fingerprinting protection (should cause problems consistently, until the feature is turned off)
some sort of intermittent bug in Firefox, possibly related to Nvidia drivers / hardware (should happen inconsistently, and may resolve itself by restarting FIrefox)
I used Edge (ugh) for about a four weeks (admittedly in area with very bad internet), when I noticed an update to Firefox available. After running that update, this resolved itself. @dagdagdiya I’ve just done an upload through Firefox and it still working fine here. Check
Step 1: Click on the species name box for help identifying an unknown species.
Step 2: Only the most generic animal names come up for all animals. The same animals for every observation (bats, turtles, frogs, squid).
Step 3: Good suggestions including correct IDs come up for all plants.
This is goofy because just last week I was getting great suggested IDs for animals in this same location (Martinique). I first noticed this problem this afternoon. I waited 5 hours and it’s still happening.
Interesting. The odd behavior only happens on the bulk upload page. Once an observation is created, if I click on the Species Name field from the observation page, Computer Vision gives me reasonable suggestions.
FWIW I use Firefox as my daily browser on my Mac and have never experienced this. I guess my laptop doesn’t use Nvidia hardware. Which fingerprinting protection should I turn on to try and replicate it?
those two pages have two different workflows for submitting images to be evaluated by the computer vision. the Firefox problems would show up only on the web upload page.
there’s HTML canvas fingerprinting protection, but i’m guessing that’s not what erikamitchell is seeing, since the problems reported are not consistent.
the reported symptoms do seem like something similar to the other past Firefox problem though, since the resulting suggestions are things things that might result from basically feeding the computer vision a single color image or an image that is effectively static (ex. a black image might be interpreted by computer vision as bats at night).
i’m not sure that getting the original photos matters. i bet even if erikamitchell tried loading the same image that didn’t give good CV suggestions before, that it might give good CV suggestions in a different upload.
here’s an example of the suggestions that i get when uploading a completely black image (notice the bats and sea creatures and such):
i’m not sure that erikamitchell’s browser is necessarily sending the computer vision a black image to evaluate – it might be a different single color or some sort of static – but if it is a black image, then i bet if you start your test with a black image change the location and date to match whatever the location date of the problematic observations were, then the additional weighting for “expected nearby” might give you the same kinds of results that were shown in the previous images showing problems.
this was never answered directly, but it looks like the last screenshot shows enough of a centered URL box that confirms it is Firefox.
what’s your workflow for cropping photos before uploading?
do you have the latest version of firefox? i assume you have an nvidia gpu. if so, what happens if you turn off hardware graphics acceleration in the browser?
