Platform: Website
Browser: Firefox & Chrome
URLs:
Screenshots of what you are seeing:
Description of problem:
While uploading multiple observations, and after entering the dates and location of all of the photos, one observation had zero “seen nearby” suggestions for both photos. What’s weird is that the top two suggestions are a genus which, given GBIF data, is limited to Australia and SE Asia. iNaturalist does not have any other observations within 4000 km of my location. The only other observation appears to be a single, non-RG photo in Florida.
Although the screenshot is from the web uploader, suggestions on the observation page continue to make the same taxa suggestions.
I thought that the issue of suggestions of taxa from the wrong continent was previously solved, so this prompted me to wonder if something changed in the latest computer vision update.
The seen nearby filter is certainly still active and turned on for me when I check my own records. On this I too get the same results as above.
I suspect what is happening is if it finds no matches that fit the geography of the record it defaults back to simply visual match.
I’m not sure if this happened prior to the new model rollout or not though.
I just did a round of uploading and found the same problem. It suggested saltwater crocodile and a species of cat for an american alligator and a tarmandua for nine-banded armadillos among other species not found in north america
So when I upload observations with multiple photos, I initially check each photo’s suggestions separately. For whatever reason one photo might have better or worse suggestions than the rest. With the spider photos that I uploaded in that set of observations, the suggestions were very clear on what is “pretty sure” vs "not confident.
On the moth image, I had not clicked on the “Include suggestions not seen nearby” option, yet it appears to be automatically switching into that mode without clearly informing me, the user.
Also, I just checked the Help page and it mentions this:
By default, the iNaturalist only displays suggested taxa that are visually similar and have been seen nearby if visually similar taxa have been seen nearby. If no visually similar taxa have been seen nearby, it displays visually similar taxa regardless of where they have been observed.
https://www.inaturalist.org/pages/help#cv-seen-nearby
So perhaps I’m just being confused since the CV is saying “We’re pretty sure this is in the genus” when that makes no sense.
If it defaults to pure visually similarity, I’d like a way to force the suggestions to only consider “seen nearby” even if those suggestions aren’t great.
3 Likes
