Hi,
I have started to use the face detection/recognition features in digiKam 2.0.0 and now I'm wondering on how to increase the hit rate for the recognition. As it is now it's 0 %. digiKam has detected just over 4000 face and I have manually added peoples to a couple of hundred of them but still no luck with the automatic face recognition. I have played around with the parameter (Fast/Accurate) but it doesn't seem to do anything to the result. Do I have to search and merge more? Do I have to manually add tags to more faces? Both takes some time though... /Anders _______________________________________________ Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
> I have started to use the face detection/recognition features in > digiKam 2.0.0 and now I'm wondering on how to increase the hit rate > for the recognition. As it is now it's 0 %. > > digiKam has detected just over 4000 face and I have manually added > peoples to a couple of hundred of them but still no luck with the > automatic face recognition. > > I have played around with the parameter (Fast/Accurate) but it doesn't > seem to do anything to the result. Do I have to search and merge more? > Do I have to manually add tags to more faces? Both takes some time > though... Dont spend your energy - you'll not get more than 0% at the moment. We have a SoK student working on this field. It's not a trivial problem as you may imagine - good algorithms implemented in free code and applicable to real- world problems are rare. _______________________________________________ Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
In reply to this post by Anders Stedtlund
That´s the state it is in at the moment, as far as I know, never heard
of anybody having success with it. But detecting faces works great, and as it always suggests the latest applied tag for a face, you can go pritty quick through all by looking for the same person all the time, and go for the next one if finished. If the state of matters has been changed recently I would be happy to read about it. Rinus Op 21-08-11 15:49, Anders Stedtlund schreef: > Hi, > > I have started to use the face detection/recognition features in > digiKam 2.0.0 and now I'm wondering on how to increase the hit rate > for the recognition. As it is now it's 0 %. > > digiKam has detected just over 4000 face and I have manually added > peoples to a couple of hundred of them but still no luck with the > automatic face recognition. > > I have played around with the parameter (Fast/Accurate) but it doesn't > seem to do anything to the result. Do I have to search and merge more? > Do I have to manually add tags to more faces? Both takes some time > though... > > /Anders > _______________________________________________ > Digikam-users mailing list > [hidden email] > https://mail.kde.org/mailman/listinfo/digikam-users > _______________________________________________ Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
In reply to this post by Marcel Wiesweg
Hi Marcel,
Thank you for your answer! Never thougth it was an easy task to do... :) /Anders 2011/8/21 Marcel Wiesweg <[hidden email]>: > >> I have started to use the face detection/recognition features in >> digiKam 2.0.0 and now I'm wondering on how to increase the hit rate >> for the recognition. As it is now it's 0 %. >> >> digiKam has detected just over 4000 face and I have manually added >> peoples to a couple of hundred of them but still no luck with the >> automatic face recognition. >> >> I have played around with the parameter (Fast/Accurate) but it doesn't >> seem to do anything to the result. Do I have to search and merge more? >> Do I have to manually add tags to more faces? Both takes some time >> though... > > Dont spend your energy - you'll not get more than 0% at the moment. > We have a SoK student working on this field. It's not a trivial problem as you > may imagine - good algorithms implemented in free code and applicable to real- > world problems are rare. > _______________________________________________ > Digikam-users mailing list > [hidden email] > https://mail.kde.org/mailman/listinfo/digikam-users > Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
In reply to this post by Rinus
Hi Rinus,
You are correct about the last applied tag so that's a little bit easier. /Anders 2011/8/21 sleepless <[hidden email]>: > That´s the state it is in at the moment, as far as I know, never heard of > anybody having success with it. But detecting faces works great, and as it > always suggests the latest applied tag for a face, you can go pritty quick > through all by looking for the same person all the time, and go for the next > one if finished. > If the state of matters has been changed recently I would be happy to read > about it. > > Rinus > > Op 21-08-11 15:49, Anders Stedtlund schreef: >> >> Hi, >> >> I have started to use the face detection/recognition features in >> digiKam 2.0.0 and now I'm wondering on how to increase the hit rate >> for the recognition. As it is now it's 0 %. >> >> digiKam has detected just over 4000 face and I have manually added >> peoples to a couple of hundred of them but still no luck with the >> automatic face recognition. >> >> I have played around with the parameter (Fast/Accurate) but it doesn't >> seem to do anything to the result. Do I have to search and merge more? >> Do I have to manually add tags to more faces? Both takes some time >> though... >> >> /Anders >> _______________________________________________ >> Digikam-users mailing list >> [hidden email] >> https://mail.kde.org/mailman/listinfo/digikam-users >> > > _______________________________________________ > Digikam-users mailing list > [hidden email] > https://mail.kde.org/mailman/listinfo/digikam-users > Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
There's a thing that could make our tag faces work easier.
Now, when Digikam suggest that a face matches a tag, it puts that face among the others that you already confirmed as a face to that tag, and makes this without any visual distinction. Because of it, when you take a general look to a face tag, you can not distinguish faces suggested by Digikam from the ones that you manually confirmed. You need to lay you mouse over the face to identify that isn't confirmed yet. I think that digikam suggested faces unconfirmed yet must be visually marked to make them easier to identify. Anyone agrees? []'s Erick Moreno On Sun, Aug 21, 2011 at 11:28 AM, Anders Stedtlund <[hidden email]> wrote: Hi Rinus, -- Erick Moreno https://profiles.google.com/erickmoreno _______________________________________________ Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
2011/8/21 Erick Moreno <[hidden email]>:
> There's a thing that could make our tag faces work easier. > > Now, when Digikam suggest that a face matches a tag, it puts that face among > the others that you already confirmed as a face to that tag, and makes this > without any visual distinction. > > Because of it, when you take a general look to a face tag, you can not > distinguish faces suggested by Digikam from the ones that you manually > confirmed. You need to lay you mouse over the face to identify that isn't > confirmed yet. > > I think that digikam suggested faces unconfirmed yet must be visually marked > to make them easier to identify. > > Anyone agrees? Sounds fair to me. /Anders > > []'s > Erick Moreno > > > On Sun, Aug 21, 2011 at 11:28 AM, Anders Stedtlund <[hidden email]> > wrote: >> >> Hi Rinus, >> >> You are correct about the last applied tag so that's a little bit easier. >> >> /Anders >> >> 2011/8/21 sleepless <[hidden email]>: >> > That´s the state it is in at the moment, as far as I know, never heard >> > of >> > anybody having success with it. But detecting faces works great, and as >> > it >> > always suggests the latest applied tag for a face, you can go pritty >> > quick >> > through all by looking for the same person all the time, and go for the >> > next >> > one if finished. >> > If the state of matters has been changed recently I would be happy to >> > read >> > about it. >> > >> > Rinus >> > >> > Op 21-08-11 15:49, Anders Stedtlund schreef: >> >> >> >> Hi, >> >> >> >> I have started to use the face detection/recognition features in >> >> digiKam 2.0.0 and now I'm wondering on how to increase the hit rate >> >> for the recognition. As it is now it's 0 %. >> >> >> >> digiKam has detected just over 4000 face and I have manually added >> >> peoples to a couple of hundred of them but still no luck with the >> >> automatic face recognition. >> >> >> >> I have played around with the parameter (Fast/Accurate) but it doesn't >> >> seem to do anything to the result. Do I have to search and merge more? >> >> Do I have to manually add tags to more faces? Both takes some time >> >> though... >> >> >> >> /Anders >> >> _______________________________________________ >> >> Digikam-users mailing list >> >> [hidden email] >> >> https://mail.kde.org/mailman/listinfo/digikam-users >> >> >> > >> > _______________________________________________ >> > Digikam-users mailing list >> > [hidden email] >> > https://mail.kde.org/mailman/listinfo/digikam-users >> > >> _______________________________________________ >> Digikam-users mailing list >> [hidden email] >> https://mail.kde.org/mailman/listinfo/digikam-users > > > > -- > Erick Moreno > https://profiles.google.com/erickmoreno > > > _______________________________________________ > Digikam-users mailing list > [hidden email] > https://mail.kde.org/mailman/listinfo/digikam-users > > Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
In reply to this post by Erick Moreno
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Am 21.08.2011 17:34 schrieb Erick Moreno: > Now, when Digikam suggest that a face matches a tag, it puts that > face among the others that you already confirmed as a face to that > tag, and makes this without any visual distinction. Can dk already recognize/match some faces on your computer? > I think that digikam suggested faces unconfirmed yet must be > visually marked to make them easier to identify. This will be essential in order to train and improve dk's face match recognition rate. Only confirmed faces can built the basis for the recognition. Maybe this has been already implemented? Another future issue might be the changing appearance of faces within the life cycle. Different face profiles for one person (e.g. on the base of the birthday, current age) may become necessary. Ben -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.16 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iQEcBAEBAgAGBQJOu8mQAAoJEJiq7tQPRElPFBUH/31rhDz4GTX785/YR7mMhImv mVlPb2l83pN4xSHA22n4702m0cy/mARAsDhktmVvZbTyhrTKG9zcDPdr6kC3+kOR ZNssGAaE0fcyuyvN9YiFjbURjVI6oauh7E39F8sGdO0W2XDiMiK7vwcqHjchZWJ6 +SgxVOg1s/CUtR3CX12Brfr75r0RDmILz4cn5YZsMYCrZFdPMLaYFU8F6d4+YfY3 vwDWSqvWJoSAMa7DI0T+sI0Be4SHtkJkvcAf8KDT+jUJY9yGyVaDDicjKFAadwK2 AjFTykInj6AUtn40UVkIXg0kXpdZq7DElP9PzN3zTZjQZ6KNnOrYMUieosD+vlY= =wK// -----END PGP SIGNATURE----- _______________________________________________ Digikam-users mailing list [hidden email] https://mail.kde.org/mailman/listinfo/digikam-users |
Free forum by Nabble | Edit this page |