3 ms·

Incidentally I've been trying out Digikam for the last couple days, mostly because of face recognition. The whole app is pretty slow on lower end hardware, they

Incidentally I've been trying out Digikam for the last couple days, mostly because of face recognition. The whole app is pretty slow on lower end hardware, they even recommend having an SSD which my travel laptop which I tested it on doesn't have. That's when I thought I should look for a web based solution because that would also solve acessing and managing the lib from multiple devices, and now there's this post here. I guess before making another switch I should check out other solutions too and compare their feature matrix and how well those work. With digikam I noticed that it seems to overfit once you tagged a couple hundred faces of the same person, for example, and is less likely to identify that person in new photos.

ape4 6y ago
I am also trying digiKam this week. Sadly the face recognition is unusable. For example I tagged one photo with Name1 and many other photos with other names. But it always suggests Name1.
- iforgotpassword 6y ago
  Yes, I found tagging a person less than about ten times tends to mess up the recognition. The sweet spot is about 20-50, then as described above as soon as you have a couple hundred for one person it seems to overfit. I could identify half the detected faces pretty easily by re-running detection two or three times but eventually I ended up with a remaining set of faces that always got detected as the wrong person, if at all, and it's not that they are all blurry or otherwise low quality. Also I wonder how bad it is to accidentally assign the wrong tag to a person. I noticed that it happened three or four times, so I ended up having maybe 400 proper assignments and then one false tag. Could they even add a function to detect this? The next thing for me would be to research which management tools support face recognition and what they use to do it. Anything using what digikam uses would not be worth trying to me.

ape4 6y ago

I am also trying digiKam this week. Sadly the face recognition is unusable. For example I tagged one photo with Name1 and many other photos with other names. But it always suggests Name1.

iforgotpassword 6y ago
Yes, I found tagging a person less than about ten times tends to mess up the recognition. The sweet spot is about 20-50, then as described above as soon as you have a couple hundred for one person it seems to overfit. I could identify half the detected faces pretty easily by re-running detection two or three times but eventually I ended up with a remaining set of faces that always got detected as the wrong person, if at all, and it's not that they are all blurry or otherwise low quality. Also I wonder how bad it is to accidentally assign the wrong tag to a person. I noticed that it happened three or four times, so I ended up having maybe 400 proper assignments and then one false tag. Could they even add a function to detect this? The next thing for me would be to research which management tools support face recognition and what they use to do it. Anything using what digikam uses would not be worth trying to me.