Bug 434603

Similar thing happen in preview tab, there are 4 people faces, i type a name once or twice and it jump to another photo instead allow me to finish naming the 3rd and 4rd one

also it's hard to name photo that have multiple faces and are close to each other, maybe when we click the Who is this, it also change the box color so we know which face are we naming em,

and maybe hide the other box when we click the Who is this, instead have to choose the option "Show face tag" to hide em since sometime the Confirm and Remove button obscure the faces

also in the first day i'm using the Detect Faces, it seem it have issue adding the Face box (it detect a face but the face box are in the chest area instead of the face)

Sorry i don't know how to edit my comment

"also in the first day i'm using the Detect Faces, it seem it have issue adding the Face box IN A PHOTO THAT ARE SIDE WAY (ROTATED WRONG) (it detect a face but the face box are in the chest area instead of the face)"

These multi bug reports are difficult for us to act as all of your requests are already included in other bug reports. You shouldn't be working with the preview when you are in the people view. The facial thumbnails are only generated for a new confirmed face and only if the confirmation is made via the thumbnails overlay, not in the preview. You work with faces in the People View, not with images. It is normal when you preview an unconfirmed face to confirm that you jump to the next image. Since the Peoble View continues to the next face and that can also be another image. The sorting is done according to the names of the faces. We also have requests for object detection in other bug reports. So many of your problems arise from an incorrect workflow with digiKam. Maybe we should give more direction, but other users want the many options because they have built their own workflow over the years.

Maik

ah sorry this is my first time reporting a bug in here and as u guess it my first time using digikam too, 

so what you said is u always work in People view (thumbs view) not in preview mode?

since i sometime go to the preview mode to fix the rectangle box so it can cover the face correctly (since i like i have said, the first day it seem the face detection have issue with photo that are wrongly rotated (sideway) that it recognize a face but put the box not on their face (it could also be that because some photo have a picasa face tag on it since i didn't see it occurring again. 

and in the People View you cannot use the dropdown name/tag list in the Who is this box so u have to use the keyboard to type the name, even in the preview section u need click the dropdown once and close it again and click the dropdown again so it can show the whole list of tag/name. or else it will just show a single name.

and i think what make it confusing about the jump after you name the face, is the randomness of it, sometime you don't get a jump at all for an hour, so you can immediately name the next photo

and the other time it keep jumping every time u name a face and it's not always the next photo on the preview section, like example u start at the top of the face list on the unknown section, and u suddenly jump to the bottom part of the scrollbar. which mean u need to wait for the jump before u can name another face and make it harder to fix a mistake like u mistake naming them.

regarding the sort name yes it is sorted alphabetically but not when there are new unconfirmed faces, then it will sort based on the number of the unconfirmed face, which is fine if there are few new faces, but not when you just starting the face recognized and there are tons new faces in different name


and yes a direction or a video how to correctly use the face detection or face recognition and how the workflow should be done is always helpful since the documentation is a bit lackluster

I think there should be a detailed explanation to improve the workflow and the face detection and face recognizing AI, like

1. is it better to correct the name in unconfirmed face or is it better to reject the suggestion or there are no difference

2. do photo that already have face data from picasa or mylio or any other software can improve the face detection/face recognition AI?

3. is better to confirm face that blurry or is better to do the clear on first since i read somewhere here that the face detection only use 100 recent face data from each name. or is it better not to add any blurry face at all

4. is it better to add manually the box when the face is not detected or is better to use the face detection again with the skip image that already scanned disable? or is that option will result in double face box?

5. is it better to correct the box when they only detect half of the face or is better use the X (this is not the face) option

@geekguy22@gmail.com

Outside the fact that this entry is just the hell to manage as on Bugzilla entry == one bug or feature and no more, we need to have a fresh feedback about the status of all point listed in this file concerning the last 8.0.0 release just published.

Please review and list clearly the pending point to fix or to implement using the last stable release.

Gilles Caulier

@geekguy22@gmail.com,

Please take a care to my last comment, and review all points listed here with the new digiKam 8.2.0 pre-release Windows
installer available at usual place:

https://files.kde.org/digikam/

This new bundle is based on last Qt framework 5.15.11 and KDE framework 5.110.

Thanks in advance

Gilles Caulier

I am planning on rewriting the facial recognition documentation as part of the upcoming 8.5.0 release.  There will be a section on best practices for "How to train your digiKam".  I hope this will help answer your questions.

Cheers,
Mike

digiKam 8.5.0 is out with many improvements in face detection and recognition. Please update these entry accordingly with this version. Thanks in advance...

https://www.digikam.org/news/2024-11-16-8.5.0_release_announcement/

(In reply to Michael Miller from comment #8)
> I am planning on rewriting the facial recognition documentation as part of
> the upcoming 8.5.0 release.  There will be a section on best practices for
> "How to train your digiKam".  I hope this will help answer your questions.
> 
> Cheers,
> Mike

Hi Mike,
I would be interested to know what we should do in cases where a person is wearing face paint or where one of those "funny face filters" has been applied. DigiKam did recognise some of these, but I'm not even sure what to do with the ones that aren't recognised. My concern is that these images will pollute the training data if I include them.

> Hi Mike,
> I would be interested to know what we should do in cases where a person is
> wearing face paint or where one of those "funny face filters" has been
> applied. DigiKam did recognise some of these, but I'm not even sure what to
> do with the ones that aren't recognised. My concern is that these images
> will pollute the training data if I include them.

Hi Craig,
That's a very good question.  I know face paint doesn't affect face recognition, but face filters that distort the facial features is a use case I haven't tested.  I'll make sure I add this use case to our test matrix.  I'll let you know as soon as I have some results to share.

Cheers,
Mike

Created attachment 176589 [details]
attachment-2736658-0.html

Thanks.

On Sat, 14 Dec 2024 at 05:41, Michael Miller <bugzilla_noreply@kde.org>
wrote:

> https://bugs.kde.org/show_bug.cgi?id=434603
>
> Michael Miller <michael_miller@msn.com> changed:
>
>            What    |Removed                     |Added
>
> ----------------------------------------------------------------------------
>                  CC|                            |michael_miller@msn.com
>
> --- Comment #11 from Michael Miller <michael_miller@msn.com> ---
> > Hi Mike,
> > I would be interested to know what we should do in cases where a person
> is
> > wearing face paint or where one of those "funny face filters" has been
> > applied. DigiKam did recognise some of these, but I'm not even sure what
> to
> > do with the ones that aren't recognised. My concern is that these images
> > will pollute the training data if I include them.
>
> Hi Craig,
> That's a very good question.  I know face paint doesn't affect face
> recognition, but face filters that distort the facial features is a use
> case I
> haven't tested.  I'll make sure I add this use case to our test matrix.
> I'll
> let you know as soon as I have some results to share.
>
> Cheers,
> Mike
>
> --
> You are receiving this mail because:
> You are on the CC list for the bug.

(In reply to Craig from comment #12)
> Created attachment 176589 [details]
> attachment-2736658-0.html
> 
> Thanks.
> 
> On Sat, 14 Dec 2024 at 05:41, Michael Miller <bugzilla_noreply@kde.org>
> wrote:
> 
> > https://bugs.kde.org/show_bug.cgi?id=434603
> >
> > Michael Miller <michael_miller@msn.com> changed:
> >
> >            What    |Removed                     |Added
> >
> > ----------------------------------------------------------------------------
> >                  CC|                            |michael_miller@msn.com
> >
> > --- Comment #11 from Michael Miller <michael_miller@msn.com> ---
> > > Hi Mike,
> > > I would be interested to know what we should do in cases where a person
> > is
> > > wearing face paint or where one of those "funny face filters" has been
> > > applied. DigiKam did recognise some of these, but I'm not even sure what
> > to
> > > do with the ones that aren't recognised. My concern is that these images
> > > will pollute the training data if I include them.
> >
> > Hi Craig,
> > That's a very good question.  I know face paint doesn't affect face
> > recognition, but face filters that distort the facial features is a use
> > case I
> > haven't tested.  I'll make sure I add this use case to our test matrix.
> > I'll
> > let you know as soon as I have some results to share.
> >
> > Cheers,
> > Mike
> >
> > --
> > You are receiving this mail because:
> > You are on the CC list for the bug.

Hi Craig,
I was able to test this, and yes, face distortion filters have a significant impact on face recogntion to the point where most distorted faces are not recognized as the same person.  It seems that some distorted faces are modified in a similar way so that the face recognition engine (face classifier) starts to see the distorted faces of different people as the same person.  I'm not sure what we can really do to fix this in digiKam.

Cheers,
Mike

Hi,

The 8.7.0 pre-release Windows installer from today have been rebuilt from
scratch with Qt 6.8.3, KDE 6.12, OpenCV 4.11 + CUDA support, Exiv2 0.28.5, ExifTool 13.27, ffmpeg 7, all image codecs updated to last version (jxl, avif, heif, aom, etc.).

Please try with this version to see if your problem still reproducible...

https://files.kde.org/digikam/

Thanks in advance
Best regards

Gilles Caulier