The underlying algorithms used are simply not up to the task
yet, and will hopefully be improved over time.
I have always tagged every picture with names of the people on them. Quite
time consuming with thousands of pictures.
Hence I was extremely hopeful when LR was updated with face recognition. However, after having used it now since it
was released I must conclude that rather than saving me time it actually now
requires more time than before to tag pictures.
About 10% of faces that are of reasonable quality frontal poses are
never recognized as faces at all, and hence not presented with proposals. Add side profiles and you end up with missing
a lot of faces. (“False positives” where
arbitrary patterns in rock etc are recognized as faces also occur, but are much
less of a problem as there is much less hassle removing false “faces” than adding those not
found.) The artificial intelligence part
where the software is supposed to learn from confirmed faces and get better at
suggesting the correct names also have pretty unimpressive hit rate. Hence after a lot of time is spent on
approving and rejecting/correcting suggested names I still have to go through
the whole lot to pick up the 10% of faces that were never picked up at all. But
the idea is great so I hope Adobe will continue working on it to get it to a
more useful level.
Be the first to post a reply!