Discussion:
[KPhotoAlbum] Searching
Robert Krawitz
2018-10-18 00:41:03 UTC
Permalink
Something else that occurred to me about searching is that it could be
useful to save complex searches (named views, in essence). At
present, all searches are ad hoc. Just a bit of an idle thought,
maybe.
--
Robert Krawitz <***@alum.mit.edu>

*** MIT Engineers A Proud Tradition http://mitathletics.com ***
Member of the League for Programming Freedom -- http://ProgFree.org
Project lead for Gutenprint -- http://gimp-print.sourceforge.net

"Linux doesn't dictate how I work, I dictate how Linux works."
--Eric Crampton
Joe
2018-10-18 06:23:50 UTC
Permalink
Post by Robert Krawitz
Something else that occurred to me about searching is that it could be
useful to save complex searches (named views, in essence). At
present, all searches are ad hoc. Just a bit of an idle thought,
maybe.
While we're brainstorming ;)  : I'm lazy. I have about 7000 of our 35000
images tagged and will almost certainly never catch up. I can do around
100/per hour when I'm fresh, but I can't maintain that for very long.
That's almost 300 hours of work just to catch up! (And my collection of
pictures is way smaller than yours!)

What I'd like is some sort of AI thing that would look at all my
human-tagged images and learn how to tag the new/untagged ones for me
automatically, maybe leaving an untagged/unfinished tag on them so I'd
know they need to be reviewed.

I'm talking about something "conceptually" basic. Something like add the
tag "Deer" (which already exists and is selected on many pictures) to
any picture with deer in it. I've never looked at whatever facial
recognition stuff is already in KPA, but something like that for things
other than people's faces would be great.

In other words, I'm primarily looking for something which "recognizes"
features/objects (and, to a lesser extent, locations/settings (I don't
have GPS cameras)).

I am *not* looking for anything which "understands", "evaluates", or
"interprets" images in any way. I'd be totally happy with it if it
couldn't tell the difference between a picture of a painting of a deer,
a picture of a deer lawn sculpture, and a picture of a live deer. (And I
do care about things other than deer! ;)  )

I realize that  this might be a large undertaking and a bit far afield
from what KPA does now, but it would sure help me! And it is all about
classifying/organizing pictures which is KPA's claim to fame (IMHO).

I don't actually expect anybody to take this on, but wanted to put the
idea out there anyway.

Joe
Tobias Leupold
2018-10-18 08:52:55 UTC
Permalink
Hi Joe :-)
Post by Joe
I'm talking about something "conceptually" basic. Something like add the
tag "Deer" (which already exists and is selected on many pictures) to
any picture with deer in it. I've never looked at whatever facial
recognition stuff is already in KPA, but something like that for things
other than people's faces would be great.
In other words, I'm primarily looking for something which "recognizes"
features/objects (and, to a lesser extent, locations/settings (I don't
have GPS cameras)).
I worked a lot on the face recognition back then. Meanwhile, we sadly had
to kick it out again, because 1. the Digikam guys apparently don't really
want other people to use their shared library for face recognition anymore
and 2. it just didn't work very good. At no point. It worked a bit, kind of
... but after all, it was more work to correct what was wrong than to just
do it manually.

And here, we're just talking about face recognition, not evaluation of all
kinds of things on an image.

If we had an AI that could do this work for us, the whole manual tagging
thing would be obsolete ;-)

I don't want to talk too pessimist, but I'm, pretty sure that if you aren't
Google, the MI6 or whatever, you won't even remotely be able to implement
such an AI. I personally wouldn't even be able to implement an AI that can
play XXO ;-)

No harm meant! But I'm very sure this is wide out of what we can do in the
foreseeable future.

The GPS thing in contrast can be fixed quite easy: I also don't have GPS
cameras. But when I want to geotag, I carry a simple GPS logger with me
(i-Blue 747A+, really cool thingy with a cellphone battery that lasts like
forver). Only thing you have to take care of is that the clock of your
photo is correct. Then, back home, you can read out the GPS track and
correlate the images with it and voilà: you have your geotags :-)

Cheers, Tobias
Martin Hoeller
2018-10-18 09:36:08 UTC
Permalink
Hi!
Post by Tobias Leupold
Hi Joe :-)
Post by Joe
I'm talking about something "conceptually" basic. Something like add the
tag "Deer" (which already exists and is selected on many pictures) to
any picture with deer in it. I've never looked at whatever facial
recognition stuff is already in KPA, but something like that for things
other than people's faces would be great.
In other words, I'm primarily looking for something which "recognizes"
features/objects (and, to a lesser extent, locations/settings (I don't
have GPS cameras)).
I worked a lot on the face recognition back then. Meanwhile, we sadly had
to kick it out again, because 1. the Digikam guys apparently don't really
want other people to use their shared library for face recognition anymore
and 2. it just didn't work very good. At no point. It worked a bit, kind of
... but after all, it was more work to correct what was wrong than to just
do it manually.
And here, we're just talking about face recognition, not evaluation of all
kinds of things on an image.
If we are ever going to have something like this, the IMHO best approach
would be to have some kind of API where you give a picture (and maybe
some meta-data) as input and get back some suggested tags as output.

This way the user could use an AI from google, amazon, etc. or some other
source of information, whatever it might be.

But I agree with Tobias and doubt, that we will get something like this
in the near future into KPA.

regards,
- martin
Johannes Zarl-Zierl
2018-10-18 19:33:38 UTC
Permalink
Hi,
Post by Martin Hoeller
If we are ever going to have something like this, the IMHO best approach
would be to have some kind of API where you give a picture (and maybe
some meta-data) as input and get back some suggested tags as output.
This way the user could use an AI from google, amazon, etc. or some other
source of information, whatever it might be.
This approach has the additional advantage that you can use it for other
things as well: suppose you want to generate tags via a script (maybe by
correlating image GPS/date info with a travel log that you keep, or because
you already fill your "Copyright" category from some exif info). Then it would
be handy to have kphotoalbum just "ingest" the external tagging data…

Cheers,
Johannes
Joe
2018-10-18 22:01:42 UTC
Permalink
I suspected as much, but it was worth mentioning.

I'll look into the geologger.

Thanks.

Joe
Post by Tobias Leupold
Hi Joe :-)
Post by Joe
I'm talking about something "conceptually" basic. Something like add the
tag "Deer" (which already exists and is selected on many pictures) to
any picture with deer in it. I've never looked at whatever facial
recognition stuff is already in KPA, but something like that for things
other than people's faces would be great.
In other words, I'm primarily looking for something which "recognizes"
features/objects (and, to a lesser extent, locations/settings (I don't
have GPS cameras)).
I worked a lot on the face recognition back then. Meanwhile, we sadly
had to kick it out again, because 1. the Digikam guys apparently don't
really want other people to use their shared library for face
recognition anymore and 2. it just didn't work very good. At no point.
It worked a bit, kind of ... but after all, it was more work to
correct what was wrong than to just do it manually.
And here, we're just talking about face recognition, not evaluation of
all kinds of things on an image.
If we had an AI that could do this work for us, the whole manual
tagging thing would be obsolete ;-)
I don't want to talk too pessimist, but I'm, pretty sure that if you
aren't Google, the MI6 or whatever, you won't even remotely be able to
implement such an AI. I personally wouldn't even be able to implement
an AI that can play XXO ;-)
No harm meant! But I'm very sure this is wide out of what we can do in
the foreseeable future.
The GPS thing in contrast can be fixed quite easy: I also don't have
GPS cameras. But when I want to geotag, I carry a simple GPS logger
with me (i-Blue 747A+, really cool thingy with a cellphone battery
that lasts like forver). Only thing you have to take care of is that
the clock of your photo is correct. Then, back home, you can read out
the GPS track and correlate the images with it and voilà: you have
your geotags :-)
Cheers, Tobias
_______________________________________________
KPhotoAlbum mailing list
https://mail.kdab.com/mailman/listinfo/kphotoalbum
Robert Krawitz
2018-10-18 12:18:25 UTC
Permalink
Post by Joe
Post by Robert Krawitz
Something else that occurred to me about searching is that it could be
useful to save complex searches (named views, in essence). At
present, all searches are ad hoc. Just a bit of an idle thought,
maybe.
While we're brainstorming ;)  : I'm lazy. I have about 7000 of our 35000
images tagged and will almost certainly never catch up. I can do around
100/per hour when I'm fresh, but I can't maintain that for very long.
That's almost 300 hours of work just to catch up! (And my collection of
pictures is way smaller than yours!)
One way you can do this more quickly is to use quick tags (typing a
single letter while in the viewer adds a tag to the image; you can
later go back and add the real information). I usually shoot in
bursts, rather than frequent small numbers; one football game, for
example, can result in 2500-3000 frames.

Robert Krawitz <***@alum.mit.edu>

*** MIT Engineers A Proud Tradition http://mitathletics.com ***
Member of the League for Programming Freedom -- http://ProgFree.org
Project lead for Gutenprint -- http://gimp-print.sourceforge.net

"Linux doesn't dictate how I work, I dictate how Linux works."
--Eric Crampton
Johannes Zarl-Zierl
2018-10-18 19:43:56 UTC
Permalink
Hi Robert,
Post by Robert Krawitz
One way you can do this more quickly is to use quick tags (typing a
single letter while in the viewer adds a tag to the image; you can
later go back and add the real information). I usually shoot in
bursts, rather than frequent small numbers; one football game, for
example, can result in 2500-3000 frames.
That actually touches an important problem with our UI: I doubt that anyone
actually knows every smart thing that you can do to improve your workflow.

Just to take myself as an example:
- I learnt about "Ctrl+Space" in the annotation window from reading the source
code.
- I know that quick tagging exists, but I never took the time to understand
it.
- I often use "copy from previous image", but I needed to set a shortcut to
make it usable for me (which reminds me that I should set this to "Alt+^" by
default).
- It feels like there are dozens of places where you just need to know the
right keyboard-combo to do something cool.

I should add: if anyone wants to rework the handbook or post some short
tutorials on Youtube/Vimeo/Gnumediagoblin/etc., please do!

Cheers,
Johannes
Joe
2018-10-18 22:34:22 UTC
Permalink
OK. I'll bite. What does Ctrl+Space do in the annotation window?

I just played with it for a few moments and didn't notice anything.

Documentation is often as much work as programming, but it's essential
to making a great project - even though it usually doesn't get recognized.
It's definitely a lot less fun to do than programming.

Joe
Post by Johannes Zarl-Zierl
Hi Robert,
Post by Robert Krawitz
One way you can do this more quickly is to use quick tags (typing a
single letter while in the viewer adds a tag to the image; you can
later go back and add the real information). I usually shoot in
bursts, rather than frequent small numbers; one football game, for
example, can result in 2500-3000 frames.
That actually touches an important problem with our UI: I doubt that anyone
actually knows every smart thing that you can do to improve your workflow.
- I learnt about "Ctrl+Space" in the annotation window from reading the source
code.
- I know that quick tagging exists, but I never took the time to understand
it.
- I often use "copy from previous image", but I needed to set a shortcut to
make it usable for me (which reminds me that I should set this to "Alt+^" by
default).
- It feels like there are dozens of places where you just need to know the
right keyboard-combo to do something cool.
I should add: if anyone wants to rework the handbook or post some short
tutorials on Youtube/Vimeo/Gnumediagoblin/etc., please do!
Cheers,
Johannes
_______________________________________________
KPhotoAlbum mailing list
https://mail.kdab.com/mailman/listinfo/kphotoalbum
Johannes Zarl-Zierl
2018-10-19 13:35:11 UTC
Permalink
Post by Joe
OK. I'll bite. What does Ctrl+Space do in the annotation window?
It toggles between the regular annotation dialog and a full-window view of the image...
Joe
2018-10-18 22:08:14 UTC
Permalink
Using the quick tags and selecting groups of images is the only way I go
as fast as I do.
They are great features. Tokens and Annotate Multiple.

Joe
Post by Robert Krawitz
Post by Joe
Post by Robert Krawitz
Something else that occurred to me about searching is that it could be
useful to save complex searches (named views, in essence). At
present, all searches are ad hoc. Just a bit of an idle thought,
maybe.
While we're brainstorming ;)  : I'm lazy. I have about 7000 of our 35000
images tagged and will almost certainly never catch up. I can do around
100/per hour when I'm fresh, but I can't maintain that for very long.
That's almost 300 hours of work just to catch up! (And my collection of
pictures is way smaller than yours!)
One way you can do this more quickly is to use quick tags (typing a
single letter while in the viewer adds a tag to the image; you can
later go back and add the real information). I usually shoot in
bursts, rather than frequent small numbers; one football game, for
example, can result in 2500-3000 frames.
*** MIT Engineers A Proud Tradition http://mitathletics.com ***
Member of the League for Programming Freedom -- http://ProgFree.org
Project lead for Gutenprint -- http://gimp-print.sourceforge.net
"Linux doesn't dictate how I work, I dictate how Linux works."
--Eric Crampton
Loading...