Why does the suggestion algorithm give different results?

I’ve noticed the suggestion algorithm sometimes gives different results on the upload page compared to the observation page (once the photos are uploaded). I don’t know if this is new behavior, perhaps I wasn’t paying attention before. Has anyone else noticed this? Is this expected?

Here’s a screen shot of an upload in progress:

On the upload page, the suggestion algorithm is focused on a plant disease, not the plant itself. The suggestion may be correct, I don’t know, but in any case I overrode the suggestion algorithm by identifying the plant instead. On the resulting observation page, the suggestion algorithm is focused on the plant, not the disease. You can confirm that by clicking on the link.

I duplicated the observation for the plant disease. On the duplicate observation page, the suggestion algorithm is focused on the plant disease, just like the upload page. Again, you can confirm that by clicking on the link.

So it seems like the suggestion algorithm on the observation page is taking the current identifications into account. Is that true? Can someone explain how this works in general?

This is exactly right. If you withdraw your observation ID, it should give you the list it gave when you were uploading. This has been discussed here:


Thanks, I missed that.

Yes, that is what I’d expect. FWIW, I’ve never had it do otherwise.

But it’s the other direction that is more interesting. If you add an ID, the algorithm may or may not present the next identifier with a different list of suggestions. Under what conditions does an ID have an effect on the suggestion algorithm?

It seems to me that if you successfully identified more than 100 observations, of the same plant in your general area, and they received RG, then the plant appearance of these reliable RGs will weigh considerably when it will come to the computer vision.

Maybe. But I think anyone adding a subsequent ID is probably going off prior knowledge/literature, so it’s not a huge issue. AI is most helpful for people IDing their own materials, as using AI to ID others’ observations is (or should) generally discouraged.

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.