3 ms·

I have all of my photos (with the exception of smartphone photos... ugh) in a nicely constructed set of folders \photos\yyyy\yyyymmmdd\ then the folder made by

I have all of my photos (with the exception of smartphone photos... ugh) in a nicely constructed set of folders \photos\yyyy\yyyymmmdd\ then the folder made by the camera, etc. I've got a small python script to generate the folders. I use Digikam[1] to do facial recognition and tagging on them. It's finally gotten to the point where it doesn't crash all the time writing metadata, and the facial recognition is acceptable. It's nowhere near as good as Picassa was, but the last release had a horrible bug that sometimes swapped face tags if there were multiple people in a photo. Those would them lead to bad training data, and it kept getting worse. At one point in time, I JPEG2000 compressed all my photos to save disk space.... it was a huge mistake, and now I've got to try to recover the original jpegs from CDs and old backups. [1] https://digikam.org/ https://digikam.org/

jacknews 3y ago
Do any of these face recognition apps work through time? I have photos of my kids (who look similar, obviously, I guess), going from age 0. The face apps all get confused, and often match younger kids, and older photos of older kids, and fail to match younger and older photos of the same kid. Do any of the apps use date-of-photo to help guide this?
- mikewarot 3y ago
  My daughter and niece look almost alike, but separated by a decade. I keep a watch out for mistakes, but for the most part, Digikam seems to do a good job. I've not noticed significant trouble recognizing faces as people age. I am fairly certain that the dates on photos are NOT used.
- Helmut10001 3y ago
  Photoview (self-hosted web app) does a pretty good job to group pictures by faces. It first self-detects similar faces and then lets you merge these groups together. New pictures then get directly assigned to these groups. For kids, I usually had pretty large self-detected groups, e.g. years <1; 1-2, 1-7; 7>. The most mismatched group was baby pictures (<1 years old kids).

jacknews 3y ago

Do any of these face recognition apps work through time? I have photos of my kids (who look similar, obviously, I guess), going from age 0. The face apps all get confused, and often match younger kids, and older photos of older kids, and fail to match younger and older photos of the same kid. Do any of the apps use date-of-photo to help guide this?

mikewarot 3y ago
My daughter and niece look almost alike, but separated by a decade. I keep a watch out for mistakes, but for the most part, Digikam seems to do a good job. I've not noticed significant trouble recognizing faces as people age. I am fairly certain that the dates on photos are NOT used.
Helmut10001 3y ago
Photoview (self-hosted web app) does a pretty good job to group pictures by faces. It first self-detects similar faces and then lets you merge these groups together. New pictures then get directly assigned to these groups. For kids, I usually had pretty large self-detected groups, e.g. years <1; 1-2, 1-7; 7>. The most mismatched group was baby pictures (<1 years old kids).