Roger
If I understand - you don't believe that a "computers and software do the
work instead of people" approach to archiving a searchable video database -
is not going to happen in the for-see-able future. I suppose you see change
as happening in a step by step linear fashion - I don't believe that is
always the case.
However in the interim, if the only approach is to "hand label" video frames
using Quicktime's "architecture" - that means that a smarter approach needs
to happen. Using Salsa dancing again as an example - one needs to Simplify
(have you read Edward De Bono's Simplicity?) first - there are several
thousand Salsa "how to" and competition videos available - but instead of
archiving all the thousands of videos - instead there are only about 400
distinct Salsa moves. So just 400 moves would have to be manually labelled.
And then one could be able to 'ask for' a Cross Body Lead with an Inside
Turn etc. etc.Assuming the protocols are finished to do MPEG-4 BIFS - any
reason that the recall and/or combination of the videos can't be done once
the hand work is done?
As you probably know - the whole "terrorism" issue has caused a multi
billion dollar industry to flourish overnight.There is at least one company
that does Matching of pictures of people's faces entering a football stadium
(for example) with a database of known "terrorists" (or is it a profile of a
terrorist?) - and is supposed to be accurate. I can't see why similar
software can't be used to Find A Similar Video to This Video.
- Harry Pasternak
>
> The issue is simple, and as someone working in the humanities, within an
organization with first-class experience in massive metadata creation and
standards projects I believe I can summarize for you:
>
> The problems have NEVER been, and will NOT be anytime in the near future,
technology-addressable.
>
> Yes, once you have proper, usable, controlled metadata that has
anticipated all the use-cases of all possible interactions, then metadata
standards (such as the future MPEG standards) do have some relevance when it
comes to tying in several systems under a single data model.
>
> The real issue, and what David, Charles, and others have been reflecting
on, is that the actual process of creating the metadata is about 1000 times
more labor intensive than the efforts involved with merely implementing one
metadata technical standard or another:
>
>
> 1) you must perform an exhaustive analysis of your uses cases; this is
HIGHLY content specific, has massive cultural implications (someone in the
US searching for salsa dances may even inflect their searches differently
enough from someone in Canada, due to cultural and linguistic biases), and
is simply not something to be underestimated. Even making a relatively small
(thousands) image catalog effectively searchable requires enormous insight
into the minds of the users. Making video segments effectively searchable
requires at least an exponentially larger ammount of effort, if you are to
be enabling searching on motion attributes, such as dance steps where you
must bracket everything in time (another dimension of cataloging).
>
> 2) you must translate your findings into effective controlled
vocabularies; in other words, develop your own semantic standards for
describing all the unique aspects of your content you wish to index.
>
> 3) you must hire a lot of folks, or find a generous university with a good
library sciences program full of willing potential interns to do your
cataloging (indexing). These interns must be rigorously familiar with both
your own semantic standards, as well as the subject matter at hand. This
means massive training and qualiy control, auditing all along the way, and
testing testing testing (usually you run an initial pilot with a tiny subset
of your material, test your use cases against that, and then scale and test
repeatedly).
>
> 4) you must constantly iterate your efforts, recataloging obsoleted
content (due to changes in your metadata models to accomodate new elements
not anticipated early on), reassessing your data in light of new use cases,
etc
>
>
> NONE of this, as you can see, is yet technologically limited, nor is
particularly ripe for technology solutions at this point. I am not a great
dreamer of sci-fi scenarios, so let's just say that for the near future,
there is no conceivable or even plausible solution to these generalized
problems. Computers are simply not good at analyzing content for indexing,
beyond simple (and unfortunately over-hyped) applications such as
transcriptions of certain TV programming using speech recognition.
>
> Therefore, the answer to your dreams is NOT MPEG7 or MPEG21 - it is brute
force, I am sorry to say. The choice of a particular metadata model is
trivial compared to the difficulties of even beginning to engineer the
processes I describe above; and that was hardly an exhaustive description of
a complete process - instead, I just touched on some of the big points and
their relative magnitude of effort. The choice of media architectures
(Quicktime, MPEG-4, Windows Media), database platforms, and even data
encoding is completely superficial to the majority of your task; the
generation of your content, and then the appropriate cataloging, is the hard
part - and always underestimated and underplanned for.
>
> In fact, if you dream and truly desire such capabilities in the future,
the only way you'll get there within your lifetime will likely be to begin
now; keep your metadats in a neutral format, and not even think of selecting
a final implementation platform for any of the technology components
(metadata repository, search engine, media formats, etc) until you've
accomplished 1, 2, and 3. If you plan well, you can always migrate your data
to a new standard as it is available - but waiting for that standard to be
complete until you even bother will get you exactly nowhere.
>
> If only it were as simple as MPEG-7, MPEG-21, BIFS, Intermedia, or
anything else - the problem would just be an issue of bits and wires. But
this is a human-level problem, as deep and ancient as it gets. If it were
this easy, we'd all be tapped into a massive digital repository of the
entire wealth of human knowledge, with the ability to instantly query in
natural language for anything we desire to know or experience.
>
> Unless you believe we'll all be talking to our AI-based wristwatches in 20
years, with complete contextual and semantic flexibility that even many
humans seem incapable of, then what you're looking for is not a technology
product; it is manual labor. Computers are great at processing data, but
simply terrible at creating original and useful content.
>
not post admin requests to the list. They will be ignored.
_______________________________________________
quicktime-talk mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/quicktime-talk
Do not post admin requests to the list. They will be ignored.
