SUMMARY *** When i scan whole library from zero ( https://bugs.kde.org/show_bug.cgi?id=463297 ) Some face regions appears in digikam database as face tag. Every picture which is scanned, is not used as data to train recognize model. Only new one *** STEPS TO REPRODUCE 1. Scan pictures with face regions on empty digikam database 2. Try to recognize additional faces 3. Nothing is scanned OBSERVED RESULT Nothing is scanned until i find some familar face and manually tag it. Then recognize faces is working, but imho only with these new images tagged, not with all tagged person EXPECTED RESULT Scanned region faces must be used to train model too. If is not possible do this automatically, i suggest some button to train from existing data to force rebuild "persons database" SOFTWARE/OS VERSIONS Linux/KDE Plasma: Linut Mint 21.1 (available in About System) KDE Plasma Version: KDE Frameworks Version: Qt Version: ADDITIONAL INFORMATION Using BETA 8.0 appimage from official link
Did you download the model files for neural network ? Please read well the doc here : https://docs.digikam.org/en/main_window/people_view.html If something is missing, we can fix it of course in the doc. Gilles Caulier
Note that the screencasts and screenshots visible in doc have been generated with digiKam 8.0.0 and Yolo3 DNN model. Automatic Recognition work on my computer. Gilles Caulier
This behavior is currently intended for a variety of reasons. The training database must be manually recreated in the maintenance tool. Do not expect too much, face rectangles created with other programs (Picasa) are poorly recognized by digiKam. because of a different size. Maik
Folling Maik comment 13, this file is now closed as intentional