Face recognition tweaking
Hi.
I read this tip "you can also manually assign a selection of unassigned faces and queue Missing for Facial Recognition to help it learn and assign more faces automatically.", and manually re-assigned hundreds of images from an unnamed person to named persons. I then reran refresh of FACE DETECTION, which should also trigger the "missing" in FACIAL RECOGNITION. I see no changes at all. I then re-assigned a couple of hundreds more and reran the process. I understand that there is algorithm in play, but to me the non-named assets have really obvious similarity with the named assets, hence why I'm trying to understand how I can try to tweak the operation using the parameters (or if I have misunderstood the functionality). I also have a lot of faces (in my opinion), 2,148.
I'm using the standard settings:
Facial recognition model: buffalo_l
Minimum detection score: 0,7
Maximum recognition distance: 0,5
Minimum recognized faces: 3
What are the recommended approach to try to improve the results, i.e.
- Assets on unnamed persons assign into named persons
- Less people (goes hand in hand with the previous point...)
Server running on: Ubuntu 22.04 LTS
Immich version: 1.121.0
Thanks in advance!
I read this tip "you can also manually assign a selection of unassigned faces and queue Missing for Facial Recognition to help it learn and assign more faces automatically.", and manually re-assigned hundreds of images from an unnamed person to named persons. I then reran refresh of FACE DETECTION, which should also trigger the "missing" in FACIAL RECOGNITION. I see no changes at all. I then re-assigned a couple of hundreds more and reran the process. I understand that there is algorithm in play, but to me the non-named assets have really obvious similarity with the named assets, hence why I'm trying to understand how I can try to tweak the operation using the parameters (or if I have misunderstood the functionality). I also have a lot of faces (in my opinion), 2,148.
I'm using the standard settings:
Facial recognition model: buffalo_l
Minimum detection score: 0,7
Maximum recognition distance: 0,5
Minimum recognized faces: 3
What are the recommended approach to try to improve the results, i.e.
- Assets on unnamed persons assign into named persons
- Less people (goes hand in hand with the previous point...)
Server running on: Ubuntu 22.04 LTS
Immich version: 1.121.0
Thanks in advance!
