This was a clever idea but doesn't match the tagging workflow in reality. We often need to pull up very different-looking pictures, and trying to combine vectors of previously-approved pictures with a new seed photo gives results too far away from the requested photo. Should follow the KISS principle here.
This was a clever idea but doesn't match the tagging workflow in reality. We often need to pull up very different-looking pictures, and trying to combine vectors of previously-approved pictures with a new seed photo gives results too far away from the requested photo. Should follow the KISS principle here.