Currently it is not possible to invoke a re-indexing of facial recognition data at either the global or folder/collection level. For my purposes re-indexing at the folder and collection level would be sufficient.
Reason: when a face is removed from a previously indexed image it is not possible to automatically recognize faces other than reviewing many images and manually applying a face region.
Yes, this is a function that should be available. Had tagging completed on a catalog with over 30k images in LR5 and upgraded. Went through tagging with facial recognition and was nearly done (5k left) and lost all tags but a few faces. Now, I've lost all person tags, even the ones I had in LR5.
With a large catalog 200K+ images as I have this is an essential. Additionally we need some control over the granularity over the face detection - at the moment FR is missing 30-40% of faces. We need to be able to say 'try again' and search harder!
Also re-indexing needs to be in two flavours - start from scratch on the selected images and secondly and more importantly search for faces not previously found - without removing those previously found and labelled.
Also, Lightroom found almost 30,000 faces in my photos. But it is terrible at grouping the faces together. I have many, many groups of my son in stacks of 4 or 5 photos, instead of 1 stack of hundreds of photos. Now that I have named a bunch of photos, perhaps reindexing would allow Lightroom to more intelligently create larger groups of photos of the same person?
David, it only stacks them if it's really pretty sure it's the same person. This means that you can generally name the stack without having to look inside.
If it bunched all of the guesses together, half of them would be wrong, but if you want to see that view, double click on a photo of your son and confirm them in the Similar section of the Person view. That view is a lot more relaxed about suggestions.
______________________ The Lightroom Queen - Author of the Lightroom Missing FAQ & Edit Like a Pro books.
Re-running face detection, globally and at the folder and selected photo(s) level, has to be a priority for the NEXT update. The "hit rate" for simply detecting faces in a photo (not identifying, just detecting that a face is there) is incredibly low. I'm seeing less than 50% success rate in detecting faces. Recognizing who is in each photo isn't performing well either, but failing to detect the only face in a portrait photo is a pretty big bug.
Here's a rescan work around. Try rotating your images!
You may have to rotate them back, but this seems to trigger the Face reognition process again.
I made the big mistake of deleting the face catalogue accross my entire library after my PC virtually stopped responding. Apx 50,000 faces had been detected but not labelled, and the ID/naming process was so painfuly slow I figured what harm could it be starting fresh again. Little did I realise that face recogition was a one trick pony, and once run, couldn't be re run again. I'm sure many others have been tripped up by this, and find it hard to imagine how Adobe engineers could have designed the UI with such a glaring ommision.
I've tested this on small sets of 20-30 images using bulk select/actions. Rotating once triggers a rescan which picks up some faces, rotating back to original orientation forces another rescan which picks up the rest. Not the most elegant fix, but it works for me. 🙂
You know, if I actually had the time (which I don't) it wouldn't be that hard to go into the LR6 database, find out which tables & fields are marking a photo as "skip_facial_recognition", and just unmark them en masse with an sql lite database editor. But, this should be Adobe's job. Thanks for the tip!
I was amazed not to find this re-scan functionality in the tutorial video so I had a look in the manual - and did not find it. Couldn't believe it so I searched and found the discussion - and this request which I believe is really needed for a smooth use of facial recognition.
Two issues:
a) Reindexing photos - having LR clear and re-draw face regions on photos to allow people to fix mistakes (undelete, etc.)
b) Forcing LR to re-run matching logic on 1000s of faces that it guesses matches incorrectly. I've tagged 20K faces and some individuals have 100s of example faces tagged. I still have about 15K faces to go, but the match accuracy on these last images is poor (maybe 10%) instead of getting better. I want force LR to rerun the comparison logic of these labeled faces considering the faces already tagged. LR asking if my blonde niece is my dark haired son may have been acceptable at the start of the process, but I have 5K images tagged with my son, and 300 images tagged of my niece. Mismatches at this point is unacceptable.
b) Picasa offers an adjustable scale of the criteria to make a match. Such a tool could be helpful.
It would also be very nice if the facial recognition software was aware of context, because a lot of photos are already tagged with names, just not connected to the faces. If someone's name is in the keywords, chances are that the face is them, not someone that looks vaguely like them.
It would also be nice to restrict the face matching to certain groups, I'm not going to have a lot of my aikido friends show up in a photoshoot at a dance, and not a lot of dancers will show up at a family gathering, and so forth.
I did a photo shoot this weekend. Is there a way to export the files as a separate catalog, then delete all facial recognition information from that catalog, re-start facial recognition on that catalog, and reimport it to my working catalog?