[digiKam-users] Image recognition
Locked
 
|
Hello, I would appreciate an advice how to use the face recognition in 7.2.0. The face detection runs successfully and I have many unknown faces in the People Tag “Unknown”. When I try to run the face recognition, there is no item in “Unconfirmed”. Until know I have to assign the faces manually, but this should work automatically in most cases, as far as I understood this functionality. Thanks and regards Hubert Zillner |
|
|
It is important that you rebuild the training database with digiKam-7.2.0. To
do this, select the checkbox for rebuilding the training data in the face detection/recognition section of the maintenance tool. A special album selection is not necessary, all currently confirmed faces are used. The face recognition works best with face rectangles, which were also detected by digiKam. Face rectangles from other programs, e.g. Picasa or manually drawn face rectangles, work poorly. The current face recognition can deliver over 90% correct results. Maik Am Montag, 5. April 2021, 22:41:57 CEST schrieb Hubert Zillner: > Hello, > > > > I would appreciate an advice how to use the face recognition in 7.2.0. > > The face detection runs successfully and I have many unknown faces in the > People Tag "Unknown". When I try to run the face recognition, there is no > item in "Unconfirmed". > > Until know I have to assign the faces manually, but this should work > automatically in most cases, as far as I understood this functionality. > > > > Thanks and regards > > Hubert Zillner |
|
|
Hello, personally, I find it very un-intuitiv that the training data has to be rebuilt with the maintenance tool. It is a good place for doing maintenance stuff all together and I would let it there. But from a user experience view: What about placing an option in the workflow dialog? Something like a checkbox which can be selected if I have committed several faces? Furthermore, would it be possible to do that automatically? For example retrain after x faces of the same person have been committed. Thinking even further: Would it be possible to retrain the network for a single person? In this case the re-training duration would also reduce. As far as I understand the functionality in the maintenance tool, it takes all recognized and committed faces and retrains all of them. My experience from my collection is that some people (~50 to 100) are often in the pictures and others (rest of ~400) appear only once a year. With best regards Eddie Am Di., 6. Apr. 2021 um 07:53 Uhr schrieb Maik Qualmann <[hidden email]>: It is important that you rebuild the training database with digiKam-7.2.0. To |
|
|
In reply to this post by Maik Qualmann
On Tue, 06 Apr 2021 07:52:39 +0200
Maik Qualmann <[hidden email]> wrote: > It is important that you rebuild the training database with > digiKam-7.2.0. To do this, select the checkbox for rebuilding the > training data in the face detection/recognition section of the > maintenance tool. A special album selection is not necessary, all > currently confirmed faces are used. The face recognition works best > with face rectangles, which were also detected by digiKam. Face > rectangles from other programs, e.g. Picasa or manually drawn face > rectangles, work poorly. The current face recognition can deliver > over 90% correct results. > What are the steps for DK 7.2.0? Tools > Maintenance > detect & recognize faces and that's it? or do I need something extra? Thank you -- sknahT vyS |
|
|
I dont even use the maintenance tab for this.... maybe I am doing it wrong. My results have been good, though. I just use the left sidebar (I'm on Windows if that matters). Then I choose the People tab, and use detect or recognize faces. On Tue, Apr 6, 2021 at 10:55 AM <[hidden email]> wrote: On Tue, 06 Apr 2021 07:52:39 +0200 |
|
|
In reply to this post by Maik Qualmann
Hi Maik and others,
thank you for the suggestion. I have run the rebuild of the training data ( tools/maintenance/detect and recognize faces [Clear an rebuild all training data]) which took a lot of time (13,5 hours). After this task I see in the people tag "unknown" many many faces and if I click to one face the preset of the field is "Who is this?" and I can assign the name manually out of the people list. When I click another face the name of the previous face is preset. My expectation is however, that after a recognition run all the recognized faces are in the people tag "unconfirmed" but there is nothing at all in this tag. Also my expectation is, that I can choose a folder within an album (right mouse) and activate "scan for faces" to get all recognized faces in "unconfirmed" which I only have to confirm afterwards manually. So the question is, do I apply this function wrongly, or rather what should be the result of the whole process? By the way my operating system is Windows 10. Regards Hubert -----Ursprüngliche Nachricht----- Von: Digikam-users <[hidden email]> Im Auftrag von Maik Qualmann Gesendet: Dienstag, 6. April 2021 07:53 An: digiKam - Home Manage your photographs as a professional with the power of open source <[hidden email]> Betreff: Re: [digiKam-users] Image recognition It is important that you rebuild the training database with digiKam-7.2.0. To do this, select the checkbox for rebuilding the training data in the face detection/recognition section of the maintenance tool. A special album selection is not necessary, all currently confirmed faces are used. The face recognition works best with face rectangles, which were also detected by digiKam. Face rectangles from other programs, e.g. Picasa or manually drawn face rectangles, work poorly. The current face recognition can deliver over 90% correct results. Maik Am Montag, 5. April 2021, 22:41:57 CEST schrieb Hubert Zillner: > Hello, > > > > I would appreciate an advice how to use the face recognition in 7.2.0. > > The face detection runs successfully and I have many unknown faces in > the People Tag "Unknown". When I try to run the face recognition, > there is no item in "Unconfirmed". > > Until know I have to assign the faces manually, but this should work > automatically in most cases, as far as I understood this functionality. > > > > Thanks and regards > > Hubert Zillner |
|
|
Hello, I was not aware that I had to rebuild the training database. I have now done a maintenance --> Clean & update face db. This seems like something that will take many hours on my collection. I notice that DK seems to be only utilizing a single CPU core. I have 6 cores. Is retraining not a parallelizable task? Or is there just a checkbox I need to check in order to have it utilize all cores? Den tir. 6. apr. 2021 kl. 22.53 skrev Hubert Zillner <[hidden email]>: Hi Maik and others, |
|
Hi,
In facial recognition, what to do with the lot of the individuals with different faces, unidentifiable but to which we could give a generic name: indigenous Chinese, Indian, Peruvian or co-vacant, sportsman, tourist, worker ... or even faces of children, old people, ... Can we do it without misleading the AI engine. Thank-You Paul Christophe ----- Paul Christophe -- Sent from: http://digikam.1695700.n4.nabble.com/digikam-users-f1735189.html
Paul Christophe
|
«
Return to digikam-users
|
1 view|%1 views
| Free forum by Nabble | Edit this page |