I apologize if this has been discussed…
Can someone explain if there is a specific rule in place to upload images if users are depending on the AI for identification? Oftentimes I feel pretty confident I can ID my own image with certainty, but just as often I am eager to learn possible identifications from the suggestions generated by iNat. NOTE: I know it is not encouraged to rely on Computer Vision - but shouldn’t we assume a lot of users, in fact, do?
Keeping this in mind…Is the first image the only image being scanned by the AI? For example, if I upload images of an unfamiliar tree - the trunk, a close-up of the bark, a silhouette of the entire tree, a leaf, and a fruit - is one image more valuable than any other…and if that’s the case, how will AI learn to recognize other images as the same species? For example, if the most effective practice is to post the leaves, and everyone generally agrees to post an image of the leaves as the first image, how will the AI ever learn to recognize the bark?
OR…is there no preferred order?
As it stands, only the first image is considered. If the observation is yours, you can re-order the pictures and get the AI to tell you its opinion on each, then conflate the results together on your own.
There was a proposal to make the computer vision work on every picture, but IIRC it didn’t get a lot of votes.
In the training of the model, images are selected from observations regardless of the order.
When the model is deployed / used, it only looks at the first image.
Interesting. I only recently learned to use the Combine tool on upload - I used to upload a single image and then return to each observation to add additional images. It was very time-consuming!
I might begin uploading all my images, then allow the AI to suggest for all, consider the accuracy of the suggestions, ID the observation…and then select the first image before I utilize the Combine tool.
That’s exactly what I do. I upload all images (web). That way, if I’m not sure of the ID, I can see the AI suggestions for each image, instead of just the first. Those suggestions give me an idea where to start looking. Once I figure out what it is, I combine the images into one observation and add the ID. Then the order doesn’t really matter. I try to set the first image to the one I would prefer to see if I were verifying (Identify) the observation.
I always try not to use the AI, and if I have no idea, I will just use a higher classification
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.