Does the AI only use the first pic in the observation to ID, or if there are multiple pics, does it take them all into account?
(I must say, sometimes the system amazes me. A obs I know what it is, but not clear in photo, and the system guesses correct, or close to it. Creepy. To those developing it, keep up the good work.)
Only first one.
This has already been answered but let me add a little more. Since the AI can only evaluate one photo at a time, you probably want to take this into account when ordering the photos in a multi-photo observation. The first photo should be your best photo (“put your best foot forward”) but “best” depends on numerous criteria, not just the AI’s opinion.
As an observer, I’ve gotten into the habit of applying the AI to each and every photo before I group photos into observations and upload them to the system. This has helped me understand the AI’s strengths and weaknesses. For example, I’ve come to believe that the AI does a much better job of classifying bare-leaved plants (no flower or fruit) than I ever will.
Interesting! I would have assumed flowers or fruits were more distinctive and put them first. I may have to change the order of some of my observation photos now.
It all depends on what it was teached on, if most of observations are about leaves it will learn leaves, it is definitely very good at many species with good amount of observations, I was playing a game wih my husband while uploading photos of plants (mainly vegetative parts) about guessing an id, 90% of time it was the first suggestion. Well, it even correctly ids some moss species, charismatic species like Hylocomium splendens (though it brings some ids to almost any poor photo of a moss).
Note that trscavo is not saying that the Computer Vision Model does better on leaf-only images than it does on images with flowers and fruit. Instead trscavo was simply saying that the Computer Vision Model does way better than trscavo ever will on leaf-only images.
We haven’t done a good job of this ourselves (myself especially, in the past), but the computer vision model isn’t really “AI”. It’s been trained using machine learning. Probably best to refer to it as the “computer vision model” or something like that. :-)
bring back the identotron
I always wanted to call the model the “suggestotron”, but that didn’t gain any traction, alas.
The identotron still exists, just that accessing the identotron from an observation is not necessarily obvious. The trick is to right-click on an existing “Compare” button - not click on - and use “open in new tab” and this drops one into the identotron. The location may have to be adjusted to a broader category to access a more realistic list.
Right, good reminder. Though, the request was more … nomenclatural.