Re:Tags versus categories

[Indulis Bernsteins <indulis b au1 ibm com>]

>> I would like to ask whether there is really
a need to differentiate
>> between tags and categories, and if not, if the two could be merged.

I agree entirely.  In fact, take it one step
further.  A filename for an image is just another tag.  

This means if you move the files, then you can "bulk
update" the tags.  A listing of images in a directory becomes
just another example of a sort-by-tag.  Add on some basic recognition
routines, which can pull out the image ID out of the EXIF and/or recognise
the MD5sums of 2 images are the same, and you now have the ability to sort
and see a stack of related images, which are derived from the same original
photo.  T

his is already being done in a non-generic way with
the "stacking" of the original image and the altered image in
F-Spot.  It just needs some rethinking to make it general purpose,
which opens up whole new ways of managing photos (i.e. see all of the images
that were derived from one photo, delete the ones that are duplicates in
size, but leave the newest one).

The approach to having a general "tag-driven"
display interface would be v useful.  Then we can write plugins to
do things like recognise faces, and autotag them, then f-spot can automagically
adapt to these new functions.

I also think it'd be good to have tags of different
types, user-input and system generated. User-input tags should be able
to be exported between versions.  Noone wants to have to reenter 200+
tags by hand.  The derived ones can just be redone.  If the EXIF
image metadata is used to recognise an image (rather than its filename),
then you can automatically reattach the user-input tags, even if the file
locations, or even filenames have changed.  Until there is an easy
way to reattach user-input tags even when the database schema changes,
f-spot won't get its full use, as there is a big disincentive to spend
time setting up your photos, only to have a future version "de-tag"
them.

Also, user-input tags that apply to one of the images
derived from a photo can be automatically attached to all other images
derived from the same photo.  MD5sum *could* be used but IMHO it is
overkill in most cases, as most people will keep the EXIF tags. If people
want, have a background task that does MD5sums and other CPU and I/O intensive
matching algorithms.  This can slowly figure out matches in the background,
both for identical and for "derived from same photo" copies.

Cheers,

Indulis