nextcloud
Manually/automatically re-run clustering on rejected faces
Describe the feature you'd like to request
Last year, I added a few thousand photos to Memories over a few days. Recognize eventually churned through most of them and did a pretty good job of recognizing faces and tagging people. However, just a few weeks ago, I discovered the Unassigned faces page and there are so many photos and faces that remain unassigned. I spent a couple of hours assigning faces to ones that Recognize had already clustered together (most of which had a name assigned to them). I then assumed that Recognize would incrementally re-run clustering on rejected faces given the new information I provided by manually assigning a few. However, this does not seem to be the case.
Describe the solution you'd like
I would like a way to have Recognize re-run clustering on rejected faces given new stimuli (such as some rejected faces being manually assigned by the user). This would help me cut down on the number of unassigned faces I have.
I did come across https://github.com/nextcloud/recognize/issues/951 which says that clustering does grab a portion of rejected faces during the clustering of new faces. However, if I am not actively adding new photos, then this does not occur.
Describe alternatives you've considered
I've tried manually running clustering using occ but, as mentioned above, this will only work with new photos that have not yet been clustered. I'm hoping for a way to make the manual assignment of faces a new stimulus for attempting to cluster previously rejected faces.
Hello @ChapterSevenSeeds thank you for taking the time to provide feedback by opening this issue! We looked into your suggestion and it looks interesting. Would you be up for developing this and submitting a pull request? We will review the code and we are also happy to give pointers. At the moment we do not have the possibility ourselves to develop this but we are happy to help and review the PR swiftly if someone wants to work on this.
Cheers Marcel
I'd also like this feature implemented, because after adding a large collection of pictures at once, many faces are unlabeled or the same person is clustered into multiple clusters depending on age. I expect better results after the clusters are manually merged.
Guys, I hate to admit it but I decided to give Immich a try and was completely sold within the first 15 minutes. I've since moved all my photos over to there and have removed Memories and Recognize from my Nextcloud instance. This feature request certainly would still be nice, but I don't think I'll ever spend some time to work on it.