[sf-lug] Fwd: What do you people use to organize your photos both locally and on the web?
jim
jim at well.com
Sun Feb 6 08:03:43 PST 2011
wow! i usually sneer at software as being too
bloated, but this stuff seems interesting.
NeXTSTEP came out with a color wheel, i can
see using that to specify colors, which would go
a long way. i visual classes are coarse, then
rock, person, dog, tree, probably as tags, could
be mapped to images.
i make it a point not to have much stuff, so
flatfiles work great for me. still and all, i
have to work with others who have tons and tons
of stuff, pics, vids, various types of docs, all
in big numbers.
the bill/dog issue comes up a lot. i dislike
the default ~/{Pictures,Videos,Documents...}
subdirectories and favor ~/{Vacations,Sports,...}
and mainly specific project names for subdirectories,
each of which has related vids, sound, image,
diagram, text, document... and other files per
the event or focus.
On Sat, 2011-02-05 at 21:02 -0800, Brian Morris wrote:
> I recently learned a little about neural net image classification
> methods.
>
> there is a test / training data set consisting of thumbnails of
> animals, and they got pretty good result asking "which animal is
> it" (at least 50 kinds to choose from, maybe more).
>
> one guy in the ML Noisbridge group tried a little facial recognition,
> says its hard,
>
> so I am not sure you could distinguish Bill as such but you could
> place him in the category "human" and your dog in the dog category OK
>
> to run the training might take a long time but if you have a decent
> GPU (say from the last couple of years) it could run 20 times faster.
> The example I saw it was like 30 days vs 24hours, was probably high
> end GPU; but people with 2008 MacBookPro have gotten 15x speedups
> almost as good.
>
> So for this method we presume you already have a folder structure
> which you are adding to or picking from.
>
> If you want to play,
>
> for unsupervised clustering the SOM toolkit which will produce a 2-d
> visualization of fuzzy boundaried many-dimension clustered groups from
> unclustered data. Given some starting point there are refinement and
> extension methods for that. This is not neural net but well studied
> algorithm that's been around a long time.
>
> In many clustering methods supervised or not, you specify how many
> clusters you want, you do not have to separate out everything in the
> world. You choose how finely grained you want it, so you can expand
> the categories as your collection grows.
>
> If you do heirarchical clustering then you can have the file system
> tree structure. Honestly I haven't thought too much about the higher
> dimensions or the cross-linking references, however I believe that
> these are equivalent.
>
> The problem for me with flat files is it gets messy, or at least hard
> to find things. Honestly though that old program I use is flat
> (sequential) except you are allowed labels and you can sort by date,
> name, type, etc.; unless otherwise you start adding keyword tags.
> However it uses its own internal db so you are free to do a file
> system folder system of your own as well independently, which it scans
> *and* includes in its db. The catalog files are not large, and the
> program is fast on old computer. At least you have those options and
> flexibility although you do most of the organizing by hand I find it
> fairly useful, but it is definitely 1990s (production, not research)
> technology - not 'smart' in any real sense. If you want to cross-file
> you can have multiple catalogs of the same stuff and you can drag and
> drop item selections between catalog windows.
>
>
>
>
> On Sat, Feb 5, 2011 at 7:58 PM, jim <jim at systemateka.com> wrote:
>
> assume some way of working directly with images
> rather than with tags. what would be the difference
> between bill and dog?
>
> They don't look the same.
>
>
> assume some distinctions. to
> use them, you're gonna have to learn them, a whole
> vocabulary that will distinguish between red balls
> and plaid blankets and will describe every kind of
> rock, pebble, leaf, scraps of paper and plastic....
>
> You can tell them apart easily enough, but fine distinctions usually
> mean you have some special interest. They don't always require names,
> I can distinguish many colors and sort them without needing names for
> them all, nor do I need descriptions only relative comparisons, some
> way of organizing or structuring the items themselves.
>
>
>
> i'd rather use tags, but i like mid-twentieth
> century technology such as flat files, which are
> simple and take little storage.
>
>
>
>
> On Sat, 2011-02-05 at 18:02 -0800, Brian Morris wrote:
> > [whoops forgot the list]
> >
> > ---------- Forwarded message ----------
> > From: Brian Morris <cymraegish at gmail.com>
> > Date: Sat, Feb 5, 2011 at 6:01 PM
> > Subject: Re: [sf-lug] What do you people use to organize
> your photos
> > both locally and on the web?
> > To: Akkana Peck <akkana at shallowsky.com>
> >
> >
> > Keywords created flat files which I don't like much, seems
> like a
> > (mid) 20th century technology.
> >
> > I have an "Asset Management" program in Mac that is ageing /
> ailing
> > but really helps a lot, I do wish there was an open source
> replacement
> > (the newer versions of this program I don't like expensive
> bloatware).
> >
> > Even with the program I stilll use file/folder system, I
> like to
> > experiment with organizational schemes, but I only use a few
> aliases
> > here and there. I have found that most people are challenged
> to think
> > clearly in two dimensions never mind three -- we mostly live
> in two
> > dimensions unless we are pilots.
> >
> > I would like to have some running software that assists me
> in
> > classifying / reclassifying things. I have some toolkits but
> barely
> > prototypes. I'd be interested in working with others on a
> hacking
> > project maybe. My desire here at this point not to gui but
> to have
> > graphical presentation of results and some command line
> features. This
> > should be scriptable / hackable. Given a current
> representation of
> > some kind, the program could perhaps present the user with
> some
> > feedback or assist in further filing.
> >
> > Certainly if I had some better tools my life in Linux would
> be
> > expanded, I would want the tools to be cross platform though
> still and
> > cloud solutions for me are out of the question.
> >
> > What I would really like (dream) is software that could work
> directly
> > with images rather than relying (exclusively) on textual
> tags. This is
> > not impossible, but for practical purposes I think it
> requires some
> > (General Purpose) GPU programming ie OpenCL, which is an
> emerging
> > technology I hope to become involved with.
> >
> > Long story short -- getting real tired of twentieth century
> (soft)
> > technology.
> >
> >
> >
> > On Sat, Feb 5, 2011 at 4:44 PM, Akkana Peck
> <akkana at shallowsky.com>
> > wrote:
> > Mikki McGee writes:
> > > After looking at what was available, I decided
> to do it
> > all "my way."
> > > I collect art pictures from museums, and pictures
> of
> > specimens of a
> > > diverse group of living things; and so I upload
> into a
> > directory called
> > > /CAMERA, and sort and edit and same into, for
> example,
> > > /Art/Legion/Statuary. Or
> > into /Arthropoda/Insecta/Lepidoptera,
> > > or into /Pictures/Friends.
> >
> >
> > I tried that, but it got complicated and I gave up
> -- if I
> > have a
> > photo that has my friend Bill and his dog, do I put
> duplicate
> > copies
> > in Images/People/Bill and Images/Animals/Dogs? Put
> the pic in
> > one
> > place and symlink to the other place?
> >
> > I organize photos in directories by year, and within
> each year
> > I
> > just make descriptive names for upload directories,
> like
> > Images/2011/RSA-baby-quail if I went on a hike at
> Rancho San
> > Antonio where I saw a lot of baby quail.
> >
> > Then each of these directories has a Keywords file
> (just a
> > text file,
> > keyword: file1.jpg file2.jpg ...) and I have a
> script that can
> > search recursively for keywords.
> >
> > I know, you're probably thinking, "What a lot of
> wasted
> > effort!
> > [insert favorite big bloated Gnome app or
> proprietary app] can
> > do
> > all that and has a GUI too!" And probably you're
> right. But
> > with
> > my way, I can change my filing scheme or the way I
> access
> > keywords
> > at any time, I can copy any subset of my images to
> another
> > machine
> > (any platform) at any time, and I never have to
> worry about
> > how
> > to migrate a database if the program ever stops
> being
> > maintained or
> > changes its UI in a way I don't like. There are some
> > advantages to
> > the old-school text file approach.
> >
> > ...Akkana
> >
> > _______________________________________________
> >
> > sf-lug mailing list
> > sf-lug at linuxmafia.com
> > http://linuxmafia.com/mailman/listinfo/sf-lug
> > Information about SF-LUG is at
> http://www.sf-lug.org/
> >
> >
> >
> >
> > _______________________________________________
> > sf-lug mailing list
> > sf-lug at linuxmafia.com
> > http://linuxmafia.com/mailman/listinfo/sf-lug
> > Information about SF-LUG is at http://www.sf-lug.org/
>
>
>
> _______________________________________________
> sf-lug mailing list
> sf-lug at linuxmafia.com
> http://linuxmafia.com/mailman/listinfo/sf-lug
> Information about SF-LUG is at http://www.sf-lug.org/
More information about the sf-lug
mailing list