Face detection and recognition parameters

[Michael Miller]

Hi Pierre,
To what version of OpenCV digiKam is using, go to Help->Components Information->Libraries->LibOpenCV.

Cheers,
Mike

On Mar 19, 2025, at 8:22 AM, Pierre DELGOVE <dp8700 at zaclys.net> wrote:


Hi Michael,

I saw yesterday that the 8.6.0 was just released. I have to update it.

I’m using KDE Neon with KDE 6.3.

"What version of OpenCV is installed?"

I don’t know. I have to check it later (I’m not in front of this computer by now).

How do I check which version of OpenCV is used (if it is correctly installed) ?

Regards,

Pierre

Le 2025-03-19 11:02, Michael Miller a écrit :

Hi Pierre,
Please try using version 8.6.0.  It was released on 16 March. In version 8.6.0 we removed SSD, YOLO, and OpenFace from face detection.  YuNet and SFace are much more reliable.

YuNet not detecting faces is usually because of an older version of OpenCV installed on your computer.  What operating system are you using?  What version of OpenCV is installed?

Cheers,
Mike

On Mar 19, 2025, at 5:45 AM, Pierre DELGOVE <dp8700 at zaclys.net> wrote:


Hi everybody !

Due to the fact that Digikam Documentation is not quite up-to-date, I'm doing some tests using face detection, tagging and recognition in Digikam.

I have some questions about this function :

- about face detection algorithms, I've tested the three options (YuNet, YOLOv3 and the third one I don't remember about). Yunet does not detect any face although I ran it on a photo album full of faces (family photos). The different options (very small, small, medium sized, and so on) do not change the result. This algorithm is pretty useless in my opinion OR I missed something.

YOLOv3 is very efficient and detect nearly every face in the photos.

The third option provide the same results as YOLOv3.

Question : is there an option to prefer between these two algorithms ?

- about face recognition (after face tagging) and confirmation of a number of faces, there are two options in the settings : Sface and Openface.

What are the difference between these options ?

Is it possible to shift from one algorithm to another ?

What are the consequences then ?


It is said in the documentation that face recognition algorithms begin to recognise faces with a few tagged faces confirmed.

I found that one needs to confirm much more than 10-20 tagged faces to have some reliable results in the people view.


Thanks in advance for answering and helping.

Pierre



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20250319/ea295dac/attachment.htm>