thanks for pointing this out. Actually - yes - I saw it.
Unless I overread it - it does unfortunately not answer any of my questions. I need to know how to run face dectection once more on a picture that already had face detection run before but failed to recognize all faces.
As an alternative I would like to start from scratch and remove all face detection information and run FR again on all pictures.
I would also like to check in the catalog which pictures FR had already been done for.
BTW - tagging undetected faces manually is not an option . I have 70000 pictures in the catalog and there are many different people on a lot of them ... I really hope that the current result (~60% fully and clearly visible faces detected successfully) can still be improved.
LR6 can't re-index Face Recognition. Seems, you get one shot at it... that's it! After that you have to go through every image and manually draw a face region and name. FR does seem like "work in progress"
I want to do the same thing as Uri.
Face recognition has at best a 60% kill rate in my catalogue and often skips obvious faces (while finding faces in car wheels, patterned fabrics, and grassy lawns). Drawing 60,000 bounding boxes won't be possible. It'd even be nice if Lightroom would say, "these eight pictures appear to have been taken within one second of each other so these faces in profile must be of the same person" or even, "I'll copy the manually-drawn bounding box to the other seven shots." The algorithm, however, appears to analyze each shot independently and conclude nothing from the timeline. And often it doesn't see people who are oriented in profile or are looking down.
Come to think of it, even if LR6 could show all the pictures with no facial tags I could do manual tagging in high-value cases. You can sort pictures by whether they have GPS data or not, so you should be able to sort pictures by whether they have face tags. Of course, this wouldn't help in cases with multiple people in the same shot where only one of the faces got recognized as a face in the first pass (which happens often) but it would find some of the missed faces at least.
One of the biggest head-scratchers about the face recognition function is how it doesn't have a similar user operation to the rest of the program. It has the feeling of a patch or third-party script more than as an integrated module that works the same way as the others.