[digiKam-users] Face recognition speed?

KRODER Gerhard gerdst1 at gmail.com
Fri Jul 16 17:46:55 BST 2021


Hello,

is this still true with dk 7.3? I run a faces re-detection of all images 
(~20000), now for more 2 days, and about 16% of all images have been 
processed untill now. 3200 Images in more than 48h, gives a average rate 
of 66.7 images an hour, or 1.1 per minute. (ok, few videos ar processed, 
also) There must be something wrong...? Anything to change with Setup?

i use dk7.3 tar fow win64, customized setup (appdata on external USB, as 
i did before with dk 7.2, SQLlite DB, OpenCL on)

Cheers, Gerhard.


Am 14.05.2021 um 09:43 schrieb Gilles Caulier:
> Hi,
>
> Yes, the work on all processors needs optimizations. It's planned for 
> this summer where a student must work on this topic.
>
> Also, we voluntarily disable the GPU computations with OpenCV backend 
> used in the background to perform DNN operations. It's planned to 
> reactivate it in the future.
>
> There is also optimization to introduce in the data stream to 
> database, which must be more grouped to reduce database calls.
>
> Gilles caulier
>
>
>
> Le ven. 14 mai 2021 à 09:33, bsw <Brian.S.Watson at gmail.com 
> <mailto:Brian.S.Watson at gmail.com>> a écrit :
>
>     Hi,
>
>     I'm new here.  I used Adobe Lightroom for several years
>     semi-professionally.
>     Been using Google photos more recently just for the convenience. 
>     Want to
>     remove my Google dependence so upgraded my NAS and am looking for
>     local
>     management tools.  digiKam looks great and am very happy so far,
>     but, I have
>     a question about the performance of the face scanning...
>
>     I'm running 7.2.0 on Windows on a "Intel(R) Core(TM) i7-8565U CPU
>     @ 1.80GHz
>     1.99 GHz" with 16GB of RAM.  Photo library is on a NAS LAN (not WiFi)
>     attached with about 120MB/s sustained read performance. digiKam
>     database is
>     (obviously) on local storage which is SSD.
>
>     I have read a little here about face recognition so have selected
>     the YOLO
>     v3 detection model with the default 70% sensitivity and "Work on all
>     processor cores" enabled.
>
>     I've run a few tests and "Scan for Faces" on an album takes about
>     20 seconds
>     per image.  At that rate it's going to take 65 hours to scan my
>     12,374 image
>     personal library.
>
>     Is this expected or am I doing something wrong?
>
>     Thanks,
>
>     Brian
>
>
>
>     --
>     Sent from:
>     http://digikam.1695700.n4.nabble.com/digikam-users-f1735189.html
>     <http://digikam.1695700.n4.nabble.com/digikam-users-f1735189.html>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20210716/bc2953bd/attachment.htm>


More information about the Digikam-users mailing list