First steps with the new features: face recognition and versioning

classic Classic list List threaded Threaded
10 messages Options
Reply | Threaded
Open this post in threaded view
|

First steps with the new features: face recognition and versioning

S. Burmeister
Hi!

Today I compiled digikam2 against KDE 4.4 and the libk* from trunk. So I would
like to share my first steps with the new features in case that might be
interesting to those developing them.

Face recognition:
I knew that there were new features. If I had not known I would not have come
across the face recognition since it is only a little icon in the tab bar at
the left. I would suggest to check whether face recognition was ever run and
ask the user whether he wants to start scanning the collection or not,
including a checkbox "never show this again".

In fact I found the little icon in the preview before finding the tab that
allows to scan the whole collection. So my first impression was: "why do I
have to mark faces on photos if picasa does this automatically?". The latter
also indicates my bias, I know picasa's face recognition so I expected it to
work in a similar way, i.e. start to scan and group the photos and give the
user the possibility to tell which pictures were false positives.

So I tried to add some face recognition, i.e. clicked on the small icon in the
preview of a picture.

Why is there a "show face tags" icon if the picture does not have any face
tags? would it not make more sense to show the "add face tag" if there are
none yet?

When adding a face tag I tried to name it and typed Ol <- L which opened the
light table although the cursor was in the line edit for the tag name.

I closed the light table and clicked on another image where I had to click on
the "show face tags" again in order to add one. Yet now one has to click twice
on it to get the "add face tag" icon. After restarting digikam one has to only
click once for the first time again.

If one moves the red square one gets some grey or black lines that remain on
the picture but that might be a NVIDIA graphics driver issue.

If one enters a name for the tag but does not click on the tick, the tag is
not saved. I think that entering the name should be enough.

Having a look at the people tags shows a preview of the picture the tag was
added to. Yet the bit shown as preview is not the part of the image the red
square marked, i.e. not the face.

If one opens the preview of the marked photo again the red square is gone and
clicking on "show people tags" does not seem to do anything. Where is the red
square gone? Or is it just a one off? I think I thought that "add people tag"
meant the red square including the name and tag are added to the picture. But
it only means "use the red square to pick a face". So for me the red square
was part of the tag rather than a tool to add it. What about calling it
"select face", "mark face", "mark person" or "link face to person"? Then it
would be clear that nothing is added to the picture but only a face marked as
person xy.

So after searching a bit I found that one could scan the whole collection from
within the people tag tab. Yet somehow it stops after a few images, i.e. it
created an "unknown" tag and then gets stuck at some picture.

I tried several times yet I never got more than one "unknown" tag, I thought I
should get something like "person 1", "person 2" and then mark the false
positives for each tag.

Versioning:
I found versioning when saving an image, so everything fine here, since one
does not need that feature if one does not save images. However since I did
not find a "save" item anymore in the file menu I thought that "new version"
replaces "save". Thus I was surprised to see that the digikam album view
showed both versions since I expected "new version" to "overwrite" the
original, i.e. that the album view would only show the new picture.

Thinking about it it does make sense though to be able to have multiple
visible versions a picture, e.g. for comparing them on the light table.
However I think that "overwriting" is a valid use case as well because I do
not want to see the original in the album view if I corrected some colour or
contrast in the picture or resized it. I just want to see the final version.

Thus I think that there should be a "save" item which "overwrites" the
previous version, i.e. hides it in the album view and there should be the "new
version" item which does exactly what it does now, i.e. duplicate the picture
and create a separate version of it.

The album view just showed a "?" as icon at the bottom-right of the thumbnail
for the new versions. And I think that the versioning tab is too hidden, i.e.
clicking on that little new icon or right-clicking the image should give
access to the versions or open the versioning tab. At least "revert to
original" should be available from the right-click menu. There is a toolbar
button in the image editor but the album view lacks that feature.

Within the image editor I found the following:
- the text next to the thumbnails in the "versions" tab is black even for the
dark theme, so it's black text on black background. (The same is true for the
album view's standard toolbar btw)
- if one creates a new version of the picture the new thumbnail is not added
to the "versions" tab until one clicks on another picture in the preview pane
and back.
- if one right clicks on one of the applied filters it shows a "remove filter"
pop-up yet one cannot click it, i.e. nothing happens.
- clicking on the version thumbnails does not show them in the main editor
window
- saving a new version will add that thumbnail to the bottom of the preview
pane instead of where it belongs according to the sorting.

That's it until now. I'll keep on using and testing and provide feedback if
it's wanted.

Sven
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: First steps with the new features: face recognition and versioning

Martin Klapetek
Thanks a lot for this feedback, I'll provide answers for the part concerning Versioning.

On Wed, Sep 8, 2010 at 19:51, Sven Burmeister <[hidden email]> wrote:
Hi!

Versioning:
I found versioning when saving an image, so everything fine here, since one
does not need that feature if one does not save images. However since I did
not find a "save" item anymore in the file menu I thought that "new version"
replaces "save". Thus I was surprised to see that the digikam album view
showed both versions since I expected "new version" to "overwrite" the
original, i.e. that the album view would only show the new picture.

Thinking about it it does make sense though to be able to have multiple
visible versions a picture, e.g. for comparing them on the light table.
However I think that "overwriting" is a valid use case as well because I do
not want to see the original in the album view if I corrected some colour or
contrast in the picture or resized it. I just want to see the final version. 

Thus I think that there should be a "save" item which "overwrites" the
previous version, i.e. hides it in the album view and there should be the "new
version" item which does exactly what it does now, i.e. duplicate the picture
and create a separate version of it.

There is an option for exactly this in the options menu. You can either have all of the version visible, or have only the latest version visible. But this is not fully implemented yet. It is ready though, we just need a few database queries for that and few lines of code here and there and it will be working. Default behaviour will most probably be keeping only one version visible (or more as you can create a whole tree with several branches).
 

The album view just showed a "?" as icon at the bottom-right of the thumbnail
for the new versions. And I think that the versioning tab is too hidden, i.e.
clicking on that little new icon or right-clicking the image should give
access to the versions or open the versioning tab. At least "revert to
original" should be available from the right-click menu. There is a toolbar
button in the image editor but the album view lacks that feature.

Hm, there should be an icon for now, not "?". Anyway, this will get replaced once the thing described above is done, there will be different marking of versions. About the context menu - at least the revert menu item could be there. Good idea, thanks.
 

Within the image editor I found the following:
- the text next to the thumbnails in the "versions" tab is black even for the
dark theme, so it's black text on black background. (The same is true for the
album view's standard toolbar btw)

Ok, I'll set it to theme color.
 
- if one creates a new version of the picture the new thumbnail is not added
to the "versions" tab until one clicks on another picture in the preview pane
and back.

Yes, this is work in progress.
 
- if one right clicks on one of the applied filters it shows a "remove filter"
pop-up yet one cannot click it, i.e. nothing happens.

Also work in progress. No ETA on this feature though as I'd like to make the versioning itself really stable and usable first.
 
- clicking on the version thumbnails does not show them in the main editor
window

Known issue, WIP ;)
 
- saving a new version will add that thumbnail to the bottom of the preview
pane instead of where it belongs according to the sorting.

Now I'm just not sure what you mean by that...
 

That's it until now. I'll keep on using and testing and provide feedback if
it's wanted.

Sure they are, so please do test and report.

Thanks, Marty
 

Sven
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel


_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: First steps with the new features: face recognition and versioning

Aditya Bhatt
In reply to this post by S. Burmeister
Thanks for the feedback.

Face recognition:
I knew that there were new features. If I had not known I would not have come
across the face recognition since it is only a little icon in the tab bar at
the left. I would suggest to check whether face recognition was ever run and
ask the user whether he wants to start scanning the collection or not,
including a checkbox "never show this again".


Fair enough.
 

Why is there a "show face tags" icon if the picture does not have any face
tags? would it not make more sense to show the "add face tag" if there are
none yet?


Yes, noted.
 
When adding a face tag I tried to name it and typed Ol <- L which opened the
light table although the cursor was in the line edit for the tag name.


Known issue, but I'm not really able to reproduce this. Need to fix this ;-)
 
 
If one moves the red square one gets some grey or black lines that remain on
the picture but that might be a NVIDIA graphics driver issue.


Some problem with redrawing of the red rectangles. This code is probably going to be rewritten. It's not an NVIDIA issue.

If one enters a name for the tag but does not click on the tick, the tag is
not saved. I think that entering the name should be enough.

Having a look at the people tags shows a preview of the picture the tag was
added to. Yet the bit shown as preview is not the part of the image the red
square marked, i.e. not the face.


Hmm, thanks for reporting that. I didn't know about it.
 
If one opens the preview of the marked photo again the red square is gone and
clicking on "show people tags" does not seem to do anything. Where is the red
square gone? Or is it just a one off? I think I thought that "add people tag"
meant the red square including the name and tag are added to the picture. But
it only means "use the red square to pick a face". So for me the red square
was part of the tag rather than a tool to add it. What about calling it
"select face", "mark face", "mark person" or "link face to person"? Then it
would be clear that nothing is added to the picture but only a face marked as
person xy.


Yes, that's a good idea.
 
So after searching a bit I found that one could scan the whole collection from
within the people tag tab. Yet somehow it stops after a few images, i.e. it
created an "unknown" tag and then gets stuck at some picture.

I tried several times yet I never got more than one "unknown" tag, I thought I
should get something like "person 1", "person 2" and then mark the false
positives for each tag.


This is how we do it: Clicking the "unknown" tag should show you thumbnails of all faces that have not been marked yet. They are not grouped as "person 1", "person 2", and so on.
Clicking on a person's name tag should show you thumbs of all faces that have the person's name, and later (not implemented) this will show face suggestions drawn from the "unknown" pool too.
 
We're not going to group unknown persons into unknown clusters anytime soon, because I'm not sure how to do that efficiently.

Thanks for your feedback, I'm a bit short of time at the moment (real life gets in the way), but I'll try to fix some of those things.
That usability advice is exactly what we need :-)

_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: First steps with the new features: face recognition and versioning

Marcel Wiesweg
In reply to this post by Martin Klapetek

> > - the text next to the thumbnails in the "versions" tab is black even for
> > the
> > dark theme, so it's black text on black background. (The same is true for
> > the
> > album view's standard toolbar btw)
>
> Ok, I'll set it to theme color.

We adjust the application's palette with the Theme, so normally, things should
work fine using normal colors roles. The album tree views are all painted by
Qt's default delegate, no special color settings.
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: First steps with the new features: face recognition and versioning

S. Burmeister
In reply to this post by Aditya Bhatt
Am Donnerstag, 9. September 2010, 10:01:27 schrieb Sven Burmeister:
> Ok, so I wait until the "scan collection" is fixed and try again. As I
> wrote it does not continue to scan after a few images and just stays at
> one for ever.

Apologies for sending the reply to the -users list, I forgot to tell kmail
that there is a second mailinglist in that folder now.

Currently digikam2 fails to compile here because libkface fails to compile
with the following error:

/home/kde-branch/kde/src/libkface/libkface/image.cpp: In static member
function ‘static int KFaceIface::Image::recommendedSizeForDetection()’:                                                                      
/home/kde-branch/kde/src/libkface/libkface/image.cpp:125:12: error:
‘getRecommendedImageSizeForDetection’ is not a member of ‘libface::LibFace’
[ 52%] Building CXX object libkface/CMakeFiles/kface.dir/button.o
/home/kde-branch/kde/src/libkface/libkface/image.cpp:126:1: warning: control
reaches end of non-void function                                                                                                        
make[2]: *** [libkface/CMakeFiles/kface.dir/image.o] Fehler 1

libface is updated, i.e. rev 217

Sven
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: First steps with the new features: face recognition and versioning

S. Burmeister
In reply to this post by Martin Klapetek
Am Mittwoch, 8. September 2010, 23:22:08 schrieb Martin Klapetek:
> There is an option for exactly this in the options menu. You can either
> have all of the version visible, or have only the latest version visible.
> But this is not fully implemented yet. It is ready though, we just need a
> few database queries for that and few lines of code here and there and it
> will be working. Default behaviour will most probably be keeping only one
> version visible (or more as you can create a whole tree with several
> branches).

I guess being able to only use one version fits most use cases, yet for those
using a light table (which I don't) both use cases might be true. I could ask
on the -users list for people doing so and what they would prefer.

> > - saving a new version will add that thumbnail to the bottom of the
> > preview pane instead of where it belongs according to the sorting.
>
> Now I'm just not sure what you mean by that...

In the image editor one can enable a preview pane via the view menu. The
pictures in that pane are sorted by filename I think. The pane can be arranged
vertically and horizontally, I use the vertical position. So if you have
somthing like:

picture a
picture b
picture c

and create a new version of picture a it should be shown after picture a,
i.e.:

picture a
picture av1
picture b
picture c

instead it is put at the bottom of the list.

picture a
picture b
picture c
picture av1

If you close the image editor and re-open it again, the sorting is correct.

Sven
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: First steps with the new features: face recognition and versioning

Marcel Wiesweg

>
> In the image editor one can enable a preview pane via the view menu. The
> pictures in that pane are sorted by filename I think. The pane can be
> arranged vertically and horizontally, I use the vertical position. So if
> you have somthing like:

Yes, this thumbbar is not yet ported to model/view. Any new entries is always
appended to the end.

Marcel
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: [Digikam-users] First steps with the new features: face recognition and versioning

S. Burmeister
In reply to this post by Aditya Bhatt
Am Donnerstag, 9. September 2010, 10:01:27 schrieb Sven Burmeister:
> Ok, so I wait until the "scan collection" is fixed and try again. As I
> wrote it does not continue to scan after a few images and just stays at
> one for ever.

Since there are a few faces in the "Unknown tag section I tried to name them
but a) the input field is very small, especially vertically. And b) I cannot
type anything in it, i.e. I see no cursor and typing does not add any text.

I updates libface/libkface/digikam2 to the latest svn just a few minutes ago.

Sven
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: [Digikam-users] First steps with the new features: face recognition and versioning

Marcel Wiesweg

>
> Since there are a few faces in the "Unknown tag section I tried to name
> them but a) the input field is very small, especially vertically. And b) I
> cannot type anything in it, i.e. I see no cursor and typing does not add
> any text.
>
> I updates libface/libkface/digikam2 to the latest svn just a few minutes
> ago.

You testing this is great, but dont expect everything to work right
know...it's alpha status.
I plan a typable combo box (similar to "Add new tag" on the right) with a
confirmation button and probably a reject button.
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel
Reply | Threaded
Open this post in threaded view
|

Re: First steps with the new features: face recognition and versioning

S. Burmeister
Am Freitag, 10. September 2010, 14:07:02 schrieb Marcel Wiesweg:

> > Since there are a few faces in the "Unknown tag section I tried to name
> > them but a) the input field is very small, especially vertically. And b)
> > I cannot type anything in it, i.e. I see no cursor and typing does not
> > add any text.
> >
> > I updates libface/libkface/digikam2 to the latest svn just a few minutes
> > ago.
>
> You testing this is great, but dont expect everything to work right
> know...it's alpha status.
> I plan a typable combo box (similar to "Add new tag" on the right) with a
> confirmation button and probably a reject button.

I do not expect it work, I just list what I find, I could use bug reports
instead if you like. But I can also stop until you guys say that the basics
for testing are implemented. Without being able to name the faces this is not
possible.

Sven
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel