Description of problem (please provide a set of steps we can use to replicate the issue, and make as many as you need.):
On many observations that are normally pretty obvious for the viz ID to indentify (or at least make a guess), I’ve been seeing the “we’re not confident enough to make a recommendation” text with zero suggestions more than I can remember ever having seen it before. The above example is an observation that should be really easy for the viz ID, but it gives me that message. However, once I make an initial ID, it seems to be able to figure it out fine. (I’ve refrained from adding an ID to the above-linked observation to keep it as a working example.)
The single image is of a tree with pink flowers. However discerning we humans can be for that tree, that’s not a lot of detailed info for CV. That’s why a close-up of flowers for a first image is often more informative for CV. (But I just tried a closely-cropped version of the flowers on a test Upload and CV still didn’t have a suggestion. Perhaps there are just too many pink-flowered trees!)
I agree, I think I’ve noticed this more on either very complex backgrounds or very simple backgrounds, this might reflect some change in the algorithm recently that’s giving the models trouble with either end of the spectrum.
Background is often very helpful for the model to determine which species from the context of soil or other species in the background, so I’m also thinking something more systematic is going on.
I’d generally agree, a variety of photos is best for better suggestions, but it should at least recognize the flowers (flowering plants) or at least “plant” as a category here. I mostly look at plants on iNat though, so I don’t know whether this issue is exclusive to plants or more widespread among other taxa.
While I agree with some of the comments explaining how the CV doesn’t have much to go off of based purely on a photo of a pink-flowered plant, I’ve noticed a significant shortcoming with the CV suggestions. It rarely suggests any broad taxonomic category. As a result, there is a huge hole in the pattern of observations, with some listed as “unknown” or “life” because the CV doesn’t help, or suggesting to uneducated individuals an incorrect specific ID. There should be some middle ground, where the CV suggests “plants” or “insects” as the most confident category.
I kinda understand, but I cannot remember a time when an image like this wasn’t given at least suggestions – even if there are lots of pink-flowered trees, it normally always gives a handful of suggestions for pink-flowered trees. The issue here isn’t that it’s not giving a confident recommendation, it’s that it’s giving zero results at all.
Up until a couple weeks ago(?), 99% of the times I’d see “we’re not confident enough” with zero suggestions were on images where even a human would be unsure what the focus was, or whatever was the focus was blurry or something really out there.
Additionally, with this recent spate of observations with zero viz ID suggestions, once I add even a very broad ID to the image, it will start giving what I would normally expect – sometimes even being fully confident in the species. I know that this can make sense sometimes, but it feels like this is happening way more often now and on observations that it doesn’t really make sense to happen on (how is the viz ID suddenly so unsure that observations like the one I linked are for a plant?)
(Also, I’m realizing now that this may have been unclear because observations with suggestions but no confident ones still say “we’re not confident enough[…]”, but with some extra text at the end. I’m editing the post title.)
Thanks for pointing this out. We identified a bug that might be been causing this. Hopefully you should not see an empty list of results as much. I am now seeing suggestions for this photo
