No announcement yet.

Working around Face Recognition

  • Filter
  • Time
  • Show
Clear All
new posts

  • Working around Face Recognition

    An ongoing issue: face recognition in Ultimate 2022. Face recognition works, but in a fashion that makes for a lot of extra work: in a database with over 2 million photos, covering maybe ~6000 persons, where the photos are in nested subdirectories limited to one person's photos, ACDSee Ultimate 2022 seems to want to assign at least one photo to a series of persons so that if there are, for example, 3000 photos of the same person in a series of photos shoots over time, Face Recognition will assign a few hundred photos to a few hundred other persons, most of which have no apparent relation to the correct person, while a majority are recognized correctly. This happens with Face Recognition set to "Conservative". Most of these incorrect recognitions are limited to 1-3 files labeled with the incorrect identity out of the several thousand available files, where the number of incorrect identities used varies directly with the number of files scanned so that in one typical case, out of 3000 files, Ultimate assigned 65 incorrect identities over 287 files, requiring 65 corrections for this one person's photos.

    I would vastly prefer to have Ultimate refuse to 'guess' identities and leave the vast majority of photos where face recognition has run, to be left Unnamed, which makes batch corrections very simple in the Persons mode. I might think that the Conservative setting should do that, but does not. But when Ultimate makes 'bad' name choices, the workflow is tiresomely tedious, where each incorrectly recognized group of photos has to be double-clicked, reviewed and then Renamed, and then back to the Named Screen, and the next 'bad' group opened, and so on.

    So, my questions:

    1. Are there simply inherent limits to the face recognition implementation so that I should simply eschew using it? What is the practical limit for how many faces can be recognized/labeled in a Ultimate database?

    2. What is the best way to correct a really large database with thousands of 'bad' face recognitions? Turn Face Recognition off and manually label faces? Is that even possible? I think ideally, I would like to select a set of folders and simply turn every recognized face to "Unnamed" and then manually edit the face labels.

    3. What's the optimal workflow for manually labeling faces without having the Face Recognition feature active?

    4. Is it a bad (inadvisable) idea to attempt to label every single face in every single photo a (large) database?

    Thanks for any information or suggestions. I think it's a vital future feature, to implement batch-oriented manual face re-naming, almost exactly as the Persons/Unnamed mode allows unnamed photos to be manually edited in batched groups.

    Last edited by AWH; 06-02-2022, 09:39 PM.