It appears that during automatic recognition the default for unrecognised faces is to put in one of two others from those already tagged. In my case this means that any face not recognised (whether male or female, colour or black and white image) is labelled either Chris Fearnley or Julie Burford. This happened on well over one hundred faces requiring manual correction.
Your face database was created with digiKam-4.x, we had a bug there, that all entries were duplicated over and over again. You will have to delete the face database and train again. It is enough on average 3-5 faces to assign the names. After that, the automatic face recognition will assign meaningful names. Maik
After 3 weeks of work, i finally completed the compilation of AppImage using Qt 5.11.3 + QWebkit 5.212. New 6.1.0 pre-release AppImage bundle can be found here (64 bits only for the moment) : https://files.kde.org/digikam/ Please check if this bugzilla entry still valid. Thanks in advance Gilles Caulier
I have tried it but the results are inconclusive so I think I need to delete the recognition database - how can I do that? I had the options selected as in the attached image: Work on all processor cores Clear and rebuild training data Even doing that I had some very strange results.
If you wan to be sure, remove the recognition.db sqlite file. It will be re-created automatically. Gilles Caulier
I have removed the recognition db and re-run the scan but I am still puzzled by the results. In trying to name the face images I am wondering about the blue '+' and red 'x' symbols in the dialog. The red x means "not a face" but what does the blue + mean?
7.0.0-beta1 is out with new Face Recognition algorithm based on Deep Learning/Neural Network API from OpenCV https://download.kde.org/unstable/digikam/ Please test and give us a feedback Thanks in advance Gilles Caulier
See the Neural Network example from digiKam 7.0.0-beta1 performing faces recognition with 2 items face-tagged in first. https://i.imgur.com/HHF4T9X.png 1 new item is recognized, name-tagged, highlighted with green frame, and linked in "unconfirmed" virtual tag for confirmation. Recognition work well using Deep Learning and give 97% of true positive results. I close this file now. Merry Christmas and Happy new year. Gilles Caulier