[KPhotoAlbum] image search question
halkola at gmail.com
Thu Apr 26 20:58:54 CEST 2012
I can't really help you with your problem but if you are new to tagging
images I would like to say that I started with a very detailed tagging
scheme and have now later found out that relatively few tags are enough
to find the photos I want quickly. I have ~15000 photos in the database
and with a few tags I can bring the number of possible images to such
that I can easily find the one(s) I want by browsing the photos that fit
25.4.2012 9:32, josephj at main.nc.us kirjoitti:
>> On Monday 23 April 2012 03:58:58 josephj at main.nc.us wrote:
>>> Big picture - maybe a rant ;) ...
>>> As I said earlier, I'm not super at databases, but one of the most
>>> fundamental concepts I did learn was normalization. Aside from being
>>> nice and tidy mathematically, the main point of using it was to avoid
>>> sorts of problems like this.
>>> If you're going to use tag subcategories, they should be strict. I.e.
>>> pool is under hotel a, that's the only place it should be because being
>>> subcategory implies that it is just that, a subcategory of the
>>> supercategory. If it's not a true subset, then it should just be a
>>> keyword or attribute that can be applied to any object.
>> It's interesting (though not surprising) to see how the technical
>> of the users affects their expectation of what the software would do in
>> certain situations. I am working in the field of computer graphics, and
>> data structure used to represent (sub-)categories in KPA (a directed
>> graph) is so commonplace in computer graphics that I was happy to see and
>> eager to use it in KPA. The same seems not to be true for databases, which
>> are probably more familiar with than me.
>> Aside from this, I believe that allowing subcategories to be arranged not
>> in strict subset relations is a deliberately designed and implemented
>> in KPA since a strict subset editor would have been much easier to
>> (using the editing capablities of the QTreeView class). However, the
>> behaviour of the current implementation in KPA has some ambiguities which
>> should be resolved.
>> Kind regards,
>> Markus Grabner
>> Institute for Computer Graphics and Vision
>> Graz University of Technology, Inffeldgasse 16a/II, 8010 Graz, Austria
>> WWW: http://www.icg.tugraz.at/Members/grabner
> I studied databases so long ago that the only place a relational database
> existed was in an IBM lab! Hard to imagine now. My point is that I'm
> definitely not database savvy. I have no idea what Jesper had in mind
> when he built KPhotoAlbum.
> Even though it conceptually looks like what you are trying to do should
> work, I think that using the same subcategory or tag name to mean two
> different things depending on context sounds more like AI than photo
> indexing. It's the kind of thing that you come back to after a few months
> and have to figure out all over again. That level of complexity probably
> isn't needed for what you want to do.
> The main thing I am saying is that normalization is your friend. If you
> know enough math to do anything interesting with graphics - not even
> mentioning vision (which I certainly don't), then spending an hour or two
> getting familiar with normalization will probably pay off some time in the
> In the mean time, just using keywords instead of subcategories ought to
> solve your immediate problem (regardless of whether KPhotoAlbum's search
> has a bug or two.)
> KPhotoAlbum mailing list
> KPhotoAlbum at mail.kdab.com
More information about the KPhotoAlbum