[mythtv] Commercial flagging paper in December IEEE Computer

[Steven Adeff]

On 1/3/07, buzz <davidbuzz at gmail.com> wrote:
> On 1/2/07, Søren Dalsgaard <soren at dalsgaards.dk> wrote:
> > I have noticed that on my DVB-T signal it is very often the case that the
> movie is in 16:9 and the commercials before and after the programme is in
> 4:3.
> >
> > Perhaps it could be possible to add "format change" to the edit mode (cut
> point, key frame, 1 frame, 0.5 sec, 1 sec, ....) for a start?
> >
> > Thanks,
> > Søren
> >
> >
> >
> > On 31/12/06, Doug Larrick < doug at parkercat.org> wrote:
> > >
> > > FYI:
> > >
> > > "Detecting Ads in Video Streams Using Acoustic and Visual Clues" is in
> > > the December 2006 issue of IEEE Computer magazine.  Authors Michele
> > > Covell and Shumeet Baluja, Google Research and Michael Fink, The Hebbrew
> > > University of Jerusalem.
> > >
> > > The essence of the aproach described is to build a database of ads from
> > > a body of accululated programming, under the assumption that any
> > > material shown several times that's about the right length is an ad.
> > > They initially look for repetition using audio only (lower computation
> > > cost) and verify repetition by then looking at the video.
> > >
> > > The entire paper (in slightly different form) is here:
> > > www.mangolassi.org/covell/pubs/mmsp06_ads.pdf
> > >
> Kinda just a suggestion or 2, but I think that the more approaches at
> detecting adds Myth has the better.     Once there are more than a few
> different possibly competing methods, then we add some code to myth so that
> it learns from the user/s which "ad detection" schemes work bet in who
> situations.    (think baysean learning for spam filters, but applied to
> advert skipping).    YOu just give it as many "possibly applicable" advert
> related data as possible, and then make it learn.     Actually, on reading
> that again, my description sounds a lot more like a back-propogation neural
> network than a baysean filter.  either way, it still works as a concept.
>
> As far as the learning part goes:
> It's pretty trivial to know when an "ad skip" was successful... if the ad
> skip occurs and there was no user input/action required, then it was good.
>
> If you get the user doing stuff like skipping backward, pausing, or even
> editing cut points when the system was otherwise idle (just streaming
> recordings to the user) ... then the "ad skip" was bad, and should be given
> a "lower" rating.
>
> Speaking of that.... I don't know how many times I have wished there was a
> binding/button in myth for  "the last ad skip was crap and I just want to
> undo it and watch through it without skipping" - have I missed something,
> and it's already in there?   - or is there an alternate way of getting the
> same thing?
>

doesn't the new commdetector2 code use a plugin style architecture so
other methods could be created and easily added?

-- 
Steve
Before you ask, read the FAQ!
http://www.mythtv.org/wiki/index.php/Frequently_Asked_Questions
then search the Wiki, and this list,
http://www.gossamer-threads.com/lists/mythtv/
Mailinglist etiquette -
http://www.mythtv.org/wiki/index.php/Mailing_List_etiquette