<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=us-ascii">
</head>
<body>
<div style="color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);" dir="auto">
<div dir="auto" style="font-family: -apple-system, HelveticaNeue; font-size: 16px; color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);">
I have the Same issue here and would be interested in some guideline.</div>
<div dir="auto" style="font-family: -apple-system, HelveticaNeue; font-size: 16px; color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);">
<br>
</div>
<div dir="auto" style="font-family: -apple-system, HelveticaNeue; font-size: 16px; color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);">
I have seen there is one project (suggestion?) of the Google summer of code "Project: Improve Image Quality Sorter algorithms" (<a href="https://community.kde.org/GSoC/2021/Ideas#digiKam" style="">https://community.kde.org/GSoC/2021/Ideas#digiKam</a>).</div>
<div dir="auto" style="font-family: -apple-system, HelveticaNeue; font-size: 16px; color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);">
<span style="font-size: 12pt; line-height: 1.5">Can this be used to "filter" the faces used for training to overcome this?</span><br>
</div>
<div dir="auto" style="font-family: -apple-system, HelveticaNeue; font-size: 16px; color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);">
<br>
</div>
<div dir="auto" style="font-family: -apple-system, HelveticaNeue; font-size: 16px; color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);">
Cheers,</div>
<div dir="auto" style="font-family: -apple-system, HelveticaNeue; font-size: 16px; color: rgb(33, 33, 33); background-color: rgb(255, 255, 255);">
Thomas Beckler</div>
<br>
</div>
<div id="id-66d94681-8cce-4623-883f-4dafd7c2b67b" class="ms-outlook-mobile-reference-message">
<div style="font-family: sans-serif; font-size: 10.2pt; color: rgb(0, 0, 0);"><br>
</div>
<hr style="display:inline-block;width:98%" tabindex="-1">
<div id="divRplyFwdMsg"><strong>Von:</strong> Digikam-users <digikam-users-bounces@kde.org> im Auftrag von digikam-users-request@kde.org <digikam-users-request@kde.org><br>
<strong>Gesendet:</strong> Donnerstag, 19. August 2021, 13:00<br>
<strong>An:</strong> digikam-users@kde.org<br>
<strong>Betreff:</strong> Digikam-users Digest, Vol 195, Issue 20<br>
</div>
<br>
<meta name="Generator" content="Microsoft Exchange Server">
<!-- converted from text --><style><!-- .EmailQuote { margin-left: 1pt; padding-left: 4pt; border-left: #800000 2px solid; } --></style><font size="2"><span style="font-size:11pt;">
<div class="PlainText">Send Digikam-users mailing list submissions to<br>
digikam-users@kde.org<br>
<br>
To subscribe or unsubscribe via the World Wide Web, visit<br>
<a href="https://mail.kde.org/mailman/listinfo/digikam-users">https://mail.kde.org/mailman/listinfo/digikam-users</a><br>
or, via email, send a message with subject or body 'help' to<br>
digikam-users-request@kde.org<br>
<br>
You can reach the person managing the list at<br>
digikam-users-owner@kde.org<br>
<br>
When replying, please edit your Subject line so it is more specific<br>
than "Re: Contents of Digikam-users digest..."<br>
<br>
<br>
Today's Topics:<br>
<br>
1. Face Recognition Workflow (Rhys Tyers)<br>
2. Re: Face Recognition Workflow (Travis Kelley)<br>
<br>
<br>
----------------------------------------------------------------------<br>
<br>
Message: 1<br>
Date: Wed, 18 Aug 2021 18:26:59 +0100<br>
From: Rhys Tyers <rhystyers1@gmail.com><br>
To: digikam-users@kde.org<br>
Subject: [digiKam-users] Face Recognition Workflow<br>
Message-ID:<br>
<CAEg=DQH=-dJfOY-EYii0DvuM4CPdgz_69Gmw5tX0LZC1nVBzuw@mail.gmail.com><br>
Content-Type: text/plain; charset="utf-8"<br>
<br>
Hello,<br>
<br>
What is the correct way to train the face detection model when many of<br>
the faces are poor quality images?<br>
<br>
I have tens of thousands of photos which include hundreds of people many<br>
hundreds of times each. Many of their faces in these photos are quite poor<br>
quality (obscured, dark, far away).<br>
<br>
When I start recognition it is often quite poor but once I have a few high<br>
quality faces in the model it seems to work fairly well. However as I<br>
accept more of the poor quality faces I find the recognition gets worse (I<br>
assume because dark, far way, poor quality faces look very similar). People<br>
for whom I've tagged a lot of poor quality images just start getting any<br>
face suggested for them.<br>
<br>
What is the intended workflow? The application prompts you to accept<br>
recognised faces as face tags (and then they will get used in the<br>
training), so that's what I've been doing.<br>
<br>
But then there is this in the manual:* In case of unsatisfying results it<br>
might be helpful to use Clear and rebuild all training data. One reason can<br>
be that there are too many face tags assigned to a person which shows this<br>
person in a way that doesn't really help the search algorithm, e.g. with<br>
sunglasses, blurred, unusual colors, carnival make up, dark shaded areas in<br>
the face, baby/kid/adult photographs mixed.*<br>
<br>
So that implies that I should only add high quality faces to the face tags<br>
and perhaps tag poor quality ones differently? Is there some way to mark a<br>
face as "poor quality" so it is not used for training?<br>
<br>
Thanks for any help.<br>
<br>
Rhys<br>
-------------- next part --------------<br>
An HTML attachment was scrubbed...<br>
URL: <<a href="http://mail.kde.org/pipermail/digikam-users/attachments/20210818/4a6516ed/attachment-0001.htm">http://mail.kde.org/pipermail/digikam-users/attachments/20210818/4a6516ed/attachment-0001.htm</a>><br>
<br>
------------------------------<br>
<br>
Message: 2<br>
Date: Wed, 18 Aug 2021 13:33:28 -0400<br>
From: Travis Kelley <rhatguy@gmail.com><br>
To: digiKam - Home Manage your photographs as a professional with the<br>
power of open source <digikam-users@kde.org><br>
Subject: Re: [digiKam-users] Face Recognition Workflow<br>
Message-ID:<br>
<CAC0Og8nAC_vREX3ygL9WUL4UBedLiP5qMtds0dqyDjPXJ2f9JQ@mail.gmail.com><br>
Content-Type: text/plain; charset="utf-8"<br>
<br>
I'll plus one this issue. I have the same problem, where it seems that the<br>
more you tag faces the worst the algorithms get. As Rhys mentioned, not<br>
sure if the answer is to mark faces as bad or whether it would be better to<br>
be be able to somehow specify which faces to use for training. Perhaps<br>
there is a programmatic way to detect the quality of tagged faces and<br>
select the "best" images for training? I believe there is a limit of<br>
something like the last ~100 tagged faces used for training today right?<br>
<br>
On Wed, Aug 18, 2021 at 1:27 PM Rhys Tyers <rhystyers1@gmail.com> wrote:<br>
<br>
> Hello,<br>
><br>
> What is the correct way to train the face detection model when many of<br>
> the faces are poor quality images?<br>
><br>
> I have tens of thousands of photos which include hundreds of people many<br>
> hundreds of times each. Many of their faces in these photos are quite poor<br>
> quality (obscured, dark, far away).<br>
><br>
> When I start recognition it is often quite poor but once I have a few high<br>
> quality faces in the model it seems to work fairly well. However as I<br>
> accept more of the poor quality faces I find the recognition gets worse (I<br>
> assume because dark, far way, poor quality faces look very similar). People<br>
> for whom I've tagged a lot of poor quality images just start getting any<br>
> face suggested for them.<br>
><br>
> What is the intended workflow? The application prompts you to accept<br>
> recognised faces as face tags (and then they will get used in the<br>
> training), so that's what I've been doing.<br>
><br>
> But then there is this in the manual:* In case of unsatisfying results it<br>
> might be helpful to use Clear and rebuild all training data. One reason can<br>
> be that there are too many face tags assigned to a person which shows this<br>
> person in a way that doesn't really help the search algorithm, e.g. with<br>
> sunglasses, blurred, unusual colors, carnival make up, dark shaded areas in<br>
> the face, baby/kid/adult photographs mixed.*<br>
><br>
> So that implies that I should only add high quality faces to the face tags<br>
> and perhaps tag poor quality ones differently? Is there some way to mark a<br>
> face as "poor quality" so it is not used for training?<br>
><br>
> Thanks for any help.<br>
><br>
> Rhys<br>
><br>
-------------- next part --------------<br>
An HTML attachment was scrubbed...<br>
URL: <<a href="http://mail.kde.org/pipermail/digikam-users/attachments/20210818/536e4852/attachment-0001.htm">http://mail.kde.org/pipermail/digikam-users/attachments/20210818/536e4852/attachment-0001.htm</a>><br>
<br>
------------------------------<br>
<br>
Subject: Digest Footer<br>
<br>
_______________________________________________<br>
Digikam-users mailing list<br>
Digikam-users@kde.org<br>
<a href="https://mail.kde.org/mailman/listinfo/digikam-users">https://mail.kde.org/mailman/listinfo/digikam-users</a><br>
<br>
<br>
------------------------------<br>
<br>
End of Digikam-users Digest, Vol 195, Issue 20<br>
**********************************************<br>
</div>
</span></font><br>
</div>
</body>
</html>