4 Replies Latest reply on Apr 23, 2015 3:20 AM by ianbutty

    LRCC Face recognition - best practices?

    ianbutty Level 1

      Ok so we are all new to the wonderful world of face recognition in LR.  I'm trying to work out what would be the best practices for using this.

       

      A little bit of background - I have a catalog of over 200,000 images.  In addition to portrait and wedding clients, a significant part of my work is with models and another significant part of my work is theatre photography.  I have be wanting some sort of face recognition to help with both for some time.

       

      What are your namining conventions for people? - here's mine:

       

      Ideally I would label people as "surname, firstname" so that I can keep members of a family together in "named people" display, but commas are not allowed in names.  Also the professional name of many models doesn't fit that pattern eg "Strawberry Venom" or "Cute as Sin" are to models I have worked with.

       

      I am trying to come up with a sensible naming convention at the moment it is "Surname/ Firstname" for clients, theatre folk and friends/family.  Models are still a problem, at present I am thinking of "Surname/ Firstname (model name(s))"  While I may not be able to remember the real names of models, I do usually know the names from model releases.  This naming will still permit me to filter/find them in the keyword List panel by just entering the model name.

       

      On final addition I am making to this this naming convention is the use of a hashtag suffix to the name:  #F for friends and family, #C for clients, #T for theatre/actors and #M for models.  This enables me to filter on just models, or just actors, or just friends and familiy.  Where people fall into multiple categories I add multiple hashtags.  So photos of me would be keyworded with "Butterfield/ Ian #F #T"

       

      Unknown / unidentified people.

       

      What I am not yet certain about is how to handle unknown / unidentified people.  Unidentified people fall into a number of different categories.

       

      People I don't know and I am never likely to know (Eg random strangers on the street, local tour guides on holiday, random people in the background etc)

      This group is relatively easy to deal with - that is to simply delete the face recognition, End of story.

       

      People I don't know the names of yet but I am likely to find out (Eg actors in a production for which I don't have a programme)

      For these people I am making up a unique name using the format "date/ Context-Gendernn" Eg an unknown male actor at Stockport Garrick Theatre would be named as "20150313/ SGT-M01"  Although this may appear a complex solution it has a number of advantages.  If/when I do learn the name of the individual (Eg I photograph them in a different production) it is simply a case of renaming the people keyword.  Creating a unique name and not simple assigning all unknowns to a bucket name will help the face recognition algorithms find this person without it being confused by have different faces assigned to the same name. I am also using the hashtage #U to make it easier to filter the unknown faces when I need to.

       

      People I don't know the names of and there is only a slim possibility of meeting/photographing again (Eg guests as a client weeding)

      It feels as though I out to just delete the face recognition and have done with it, and this is what I would do except for thing. Other than manually drawing face regions I have not yet found a way to get lightroom to rescan a folder for faces if you have previously deleted the face recognition.  This means that deleting face regions from a large number of people is something that cannot be easily reversed.  I might just leave these people in the "Unnamed People" category... at lease until such time as there is a way to rescan a folder or colectoin.

       

      Summary

      My practices are still evolving. But I hope these thoughts and idea will help others think through the issues and come up with solutions that work for their situation.  I am interested in hearing how other people are using the face recognition system.  Especially if anyone is aware of any 'best practices' that Adobe or anyone else has recommended.

        • 1. Re: LRCC Face recognition - best practices?
          Aruna.Basnayake

          Hi - your thoughts are very helpful as I try to formulate my own rules for my LR catalog of some 110,000 photos.  So far (prior to LR CC), I have been using hierarchical keywords to tag people, eg 'Family < smith family < John'.  It looks like LR CC does not allow hierarchies for face tagging.  Is this why you are using an additinal tag '#F'?  Thoughts?

          • 2. Re: LRCC Face recognition - best practices?
            ianbutty Level 1

            Glad it helped.

             

            Yes and no.  You can still put the people keywords into hierarchies within the keyword list - you can arrange them just like any other keywords.so you just create a "smith family" keyword and store "john smith" under it.  What you can't do is apply BOTH smith family and john smith the the same face.

             

            My use of the hash tags came about because I initially had a top level keyword for models, one for clients, one for theatre peple and one for family and firends.  Then discovered that some of the theatre folk were also clients (headshots) and what to do when a friend is also a client.  So the hash tag system means a person can be both a friend, a model, an actor as well as being a client!  (#T #C #M #F).

            • 3. Re: LRCC Face recognition - best practices?
              Aruna.Basnayake Level 1

              Hi - I now understand the rationale for using #T, #F, etc - it's a great suggestion and I will adopt the same practice with my database. 

              Back to the people keywords.  What if one has multiple people under 'Smith Family'?  For example, 'Family < Smith Family < John Smith', 'Family < Smith Family < Nancy Smith'?  It didn't seem to work.  I presume, from your comments, that I need to put all names at the top level, with '/' to separate surname and first name?

              Thanks in advance.

              • 4. Re: LRCC Face recognition - best practices?
                ianbutty Level 1

                TBH although I used a family name level in the past, I might end up dropping that now and just put all the people keywords under a top level PEOPLE keyword.  Really not sure yet.  So mine now looks like this...

                 

                PEOPLE

                     Baker/ Colin #T

                     Baker/ Tom #C

                     Smith/ Matt #T #M