Created attachment 124714 [details] Test File 01 SUMMARY Attached pictures leads to no a Face detection with the standard module, but to no detection at all with the DeepLearning algorithm. STEPS TO REPRODUCE 1. Open the attached pictures, and make it the only collection. 2. Scan the whole collection, with Deep Learning 3. OBSERVED RESULT No face detected EXPECTED RESULT Scan Collection + Deep Learning should detect and suggest people group/faces SOFTWARE/OS VERSIONS Windows: 10
Created attachment 124715 [details] Test File 02
The faces are detected on both images, even with the Windows version. Maik
I tested here under Linux, no problem : faces are detected properly. Gilles Caulier
Strange... please can you point me to the right procedure to test successfully deep learning face recog nition ? i may have my digikam misconfigured... or miss a step ! Is the DNN pre-learnt ? or is it blank and i should first manually add some people before it get used to detect other peoples ?
(i enabled the checked box that erases learning database, just above the selection of the deep learning algorith. was it a mistake ?)
Are you really using the latest Windows version of files.kde.org? With all Windows versions before, the DNN face detection did not work. Maik
Hi Maik, I'm using the 22th dec 70 beta1 release available here : https://download.kde.org/unstable/digikam/ is there a log file I could open to check myself what went wrong ? or that i could maybe send to you ? Alex
Download and start DebugView from Microsoft. Start digiKam and do a face search on the image. Post the messages that are available in DebugView. Maik
Created attachment 124725 [details] DebugView Log file please find enclosed a DebugView log of the "Scan whole collection" triggered from the people view. I will also test the "Scan Face" from a single picture.
After the log, you started face recognition and not detection. Please do it with a single image. Maik
Maik, I confirm that when I click on individual "Scan for face" from "Thumbnails" or "Preview", I do see that faces are correctly detected (empty rectangles). Then how can I trigger the DNN recognition for a single image ? Moreover, from the People / whole collection, I now understand that if I do it in a 3 phases approach, it does work : 1) Detection (= scan for faces on every picture) 2) Tag some faces 3) Recognition (= DNN) will suggest other similar faces This is acceptable although not self-explanatory for non digikam native users. It would be useful to have an option to cumulate both phases ! (= Checkboxes instead of radio buttons) I also noticed that re-launching 3) can bring me new iteratively recognized faces, do you agree ? This must be manually triggered, nothing running in background ? Note that Picasa had the possibility to run this Detection + DNN recognition in a single activation, in background, and - either once (archive folder) - or endlessly (working folder) ... it was so easy to build people database in that mode ! DNN reported new similarities in real time. I just had to confirm/reallocate and after 30 minutes the whole database was rebuilt !
Hello team, how can I trigger the DNN recognition for a single image ? When I click on individual "Scan for face" from "Thumbnails" or "Preview", I do see that faces are correctly detected (empty rectangles) but there is no name suggestion (no recognition).
With the current version, facial recognition is carried out automatically for a single image. If you don't get a name, you either haven't trained your faces enough or your set accuracy value is too high, start at 50. Maik
Alexandre, I just tested under Windows 10, and i can confirm that Clear Trained Data do nothing. Previous face detected with LBPH algorithm (DK 6.x) still here. Note : i used the DK 7.0.0 version generated tomorrow. Clear Training Data option still in People left sidebar, and is not moved to Maintenance tool. I recompile the last code version for Windows now. But just to be sure that Face detection work as expected with a null database, i removed all sqlite files and restarted digiKam. I don't touch the lead DK RC configuration file here. I started a detection on my small test collection, and Unknown faces are back as expected. So the detection process work fine. https://i.imgur.com/6B93PtG.png Gilles Caulier
i used the DK 7.0.0 version generated "tomorrow" => "yesterday" of course (:=)))
Created attachment 124865 [details] attachment-559-0.html Ok, thank you Maik. Will try at 50%.
Thank you Gilles. I will also retry the maintenance one.
Gilles, deleting the training data does not delete any detected faces, that is completely normal. Because even with these, the training data for DNN are rebuilt. The training data from the LBPH are not touched, the DNN mode uses a different DB table for the data. Maik
Ok, it's more clear now.
Maintenance tested (clean database) and working fine. Thank you Maik, you can close this ticket, Resolved.
Thanks for the feedback...