I understand facetag maps map a specific face region (rectangle or polygon on an image) to a person’s identity (name/tag).
So when I correct a facetag, is the map updated to correct the identity associated with the rectangle or polygon on that image? Or is the tag for that image changed and nothing else?
If the map is updated, then I can see myself going round in circles when trying to make corrections. For instance, imagine the face tag correction was required because AI incorrectly identified two people as being the same identity, e.g. an extreme example of this could involve identical twins, because images of them may have very similar rectangles/polygons. When the map is updated, then presumably the identity of all images associated with that rectangle/polygon are updated, right? And if this is the case, then whilst the tags for one twin may be changed and thus corrected, the tags for the other twin would be changed in error.
Then subsequently, when I discover the error and correct it, then the reverse happens and I end up with a different set of errors.
The necessary solution seems to require the map to be changed to reflect two identities. And the rectangles/polygons of the two identities have to be sufficiently different so that AI will recognise the difference.
So back to my first question, when I correct a facetag, is the map updated to correct the identity associated with the rectangle or polygon for that image? Or is the tag for that image changed and nothing else? Where can I read more about how this works please?
TIA … Greg
I am with you. For some reason, QMagie thinks my daughters are the same person. It works fine when I start tagging them as 19 and 21 year olds that they are. But then when I start tagging their baby pictures, it like thinks that they are basically the same person and can be split in half and be in the same picture. The AI algorithm, should as a rule know that one person cannot be two people in the same picture and so additional logic should follow - especially when I manually go in and correct it. But nope…
when I correct a facetag, is the map updated to correct the identity associated with the rectangle or polygon for that image? Or is the tag for that image changed and nothing else?
When you change the face tag from Person A to Person B, it only updates the tag for that specific image. It does not alter the facial feature data analyzed by the AI Core.
when I start tagging their baby pictures, it like thinks that they are basically the same person and can be split in half and be in the same picture.
Currently, our recognition model has limited infant data for training. In addition, AI Core uses a lightweight model to ensure compatibility across different devices, which may affect accuracy, especially when it comes to recognizing babies.
We understand this can be confusing, and we’re looking into ways to improve recognition for infants in future updates.
Really? Because when I started manually tagging the baby pictures, suddenly all of the already tagged (and correct) identifications got changed.
when I start tagging their baby pictures, it like thinks that they are basically the same person and can be split in half and be in the same picture.
Currently, our recognition model has limited infant data for training. In addition, AI Core uses a lightweight model to ensure compatibility across different devices, which may affect accuracy, especially when it comes to recognizing babies.
We understand this can be confusing, and we’re looking into ways to improve recognition for infants in future updates.
[/quote]
Well maybe an easy way to start is that you should be able to tell if it is an infant. Maybe easiest to explain in Pseudo code:
If ImageType = Infant Then
Dont_Update_All_Other_Images
Else
Update_All_Other_Images
End If
So don’t apply the facial tagging to other images if the image being tagged is an infant.