0 Replies Latest reply on Nov 8, 2018 4:49 AM by Mikey44MagBiker

    Facial Recognition System Performance

    Mikey44MagBiker Level 1

      This is a question about how to improve the response in LR Classic Desktop.  Unless I go through unnatural gyrations to limit the scope of action, doing basic Facial Recognition tasks in LR is almost impossible. 

      Basic parameters:

      LR Catalog = 119,523 photos in one main folder with numerous subfolders

      People Named so far = 24,191 and Unnamed = 66,798

      System basics:

      Win7/64Pro, Data storage drive 10.9TB RAID, CPU 8Core 3.61GHz, RAM 16GB

      I have a big scratch disk set up on a SSD apart from storage.

      Windows Experience Index rating 7.7 overall, Graphics 7.9, Disk Transfer 7.9

       

      Weeks ago, I started Facial Recognition for the first time ever.  It took many days to complete the progress bar ‘working’ ~ during which I just experimented with a few tags just to familiarize myself with the experience.  It finally completed and I went to work.

      If I select my whole catalog, switch into Facial/People mode, it all jumps up fairly quickly to display Named/Unnamed grid.  But, double-click any person and try to make progress reviewing and adding to the tags gets extremely slow and my display freezes, flashes all-white for various durations and sometimes locks up white and must be killed with Task Manager. Believe me I’ve let it try to recover for up to four hours.  It starts up after this rude treatment with no problems or popups for any cause.  It makes no difference whether I Optimize in grid or Facial mode.  It makes no difference whether I have Excel and/or Family Tree Maker running or have NO other programs running.

      If I select a small internal folder there is what I think is normal response.  This is a very inefficient method because any face can be in any folder so, for example, I finally find an annotated image of my great-grandmother in a ‘Legacy’ folder, the other instances of her face are not presented to stack or even to confirm.

      In some software systems {ERP Business, etc.} it is a basic tenet that Quality Control is certainly responsible for basic bugs and accuracy, but also for individual user response time, multiple user stress {not applicable here} and large database performance.  These responsibilities are tested and documented before release of the version or update. These tests combine both human operator and computer generated transactions in repeatable configurations.

      So finally I come to a thinly obfuscated question:  Does Adobe have any advice for me on how to improve my workflow and improve my ability to get quicker and more accurate tags?