<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</head>
<body>
<p>Hi Maik</p>
<p><br>
</p>
<p>thanks for the clarification, that makes the process more clear.
Below tha last result I got<br>
</p>
<ul>
<li>I have run the training database using the option "clear
unconfirmed face and analyse again" (translation from french, so
may differ from the option name)<br>
</li>
<li>I have completed the face detection on all catalogs, now I
have ~20K unknown faces (using "ignore pictures already
analyzed")<br>
</li>
<li>I have run the face recognition again , and got some new
unconfirmed faces with a good success rate, but far from the
number of faces I should have exepected from the unknown<br>
</li>
</ul>
<p>what should be the options to get a better result on recognitions
?</p>
<p>Thanks and regards</p>
<p><br>
</p>
<p>frederic</p>
<p><br>
</p>
<p><br>
</p>
<p>Le 04/02/2021 à 08:16, Maik Qualmann a écrit :</p>
<blockquote type="cite" cite="mid:1624600.1lF2ilLUJX@linux-tpgn">
<pre class="moz-quote-pre" wrap="">Face recognition is carried out on all images with "Unknown" faces. So all
faces under the virtual tag "Unknown" in the People view. Album bassierd
wouldn't make sense. You probably don't even know which albums the unknown
faces come from in order to choose the right albums.
The training database rebuild is not album based either. All known faces are
used.
Maik
Am Mittwoch, 3. Februar 2021, 12:16:31 CET schrieb frederic chaume:
</pre>
<blockquote type="cite">
<pre class="moz-quote-pre" wrap="">Hi Maik
Thanks a lot
after running the training dabase, the success rate is close to 100%.
Excellent !
now still some clarification needed.
* on the face detection menu, album selection can only be modified
when "face detection" is set, no change possible when "face
recognition is set". So should we run 1st the "face detection" (and
then select the albums) before runinng "face recognitation" ?
* still seeing few faces with 1 new , while no unconfirmed is proposed
(but not systematic)
* I have run a database training on all collections , then started a
face recognitation on all collections (just set "all collections" in
face detection without running it, hope it is ok). still on-going,
will let you know the result,
thanks
frederic
Le 01/02/2021 à 22:23, Maik Qualmann a écrit :
</pre>
<blockquote type="cite">
<pre class="moz-quote-pre" wrap="">It is important that you rebuild the training database. This is done in
the
maintenance tool under face detection/recognition activate the checkbox
for
the rebuild the training database. The settings for the albums to be
searched are in a tab in the left People sidebar.
Maik
Am Montag, 1. Februar 2021, 09:48:56 CET schrieb frederic chaume:
</pre>
<blockquote type="cite">
<pre class="moz-quote-pre" wrap="">Hi all
just installed 7.2 RC on windows and start playing with face tagging ,
below a 1st quick feedback
* 1st of all: I like the presentation and the speed. it is much more
efficient then previous release (ease to find confirmed,
unconfirmed)
* seems the number of initial tags when I start using 7.2 is much
lower than what I have in previous release . Is that normal?
* some face tag is shown with 1 new , but no new is proposed, only the
confirmed ones are available/visible
* with the 1st run , the success rate is very low (<10%), so I have
unconfirmed most of them (and confirmed the positive ones)
* 2nd run very few new face is proposed for the existing tags (1 or
2 in some faces tags)
o I have thousands of "unkonwn" but was thinking they could be
proposed to the existing faces as "unconfirmed" faces?
o how to select albums? in the people menu on the left, I see "no
album selected" and didn't find how to add it (I'm using
recognize face). I tried using the maintenance tool, but no
change
o do we have to run "detect face" before "recognize face" ? I
tried also and didn't get more new unconfirmed proposed in the
face tags
Did I miss something ?
thanks and regards
frederic
</pre>
</blockquote>
</blockquote>
</blockquote>
<pre class="moz-quote-pre" wrap="">
</pre>
</blockquote>
</body>
</html>