[mythtv-users] Re: Torrentocracy Patch against Mythtv-cvs

[Brad Templeton]

On Thu, Mar 03, 2005 at 10:50:57AM -0500, Maverick wrote:
> > Actually, where it would come in handy is not just commercials (which
> > software today can detect) but other things like pitching changes, boring
> > academy awards acceptance speeches, long driving scenes to no purpose etc.
> 
> That's cool. Of course implementing this using a central server of
> some sort will be easier. Is the MythTV project (users) opposed to
> server centric information sharing, such as flagged commercials?

I think the approach is fairly straightforward.   While a user is
watching a program, you would build a map.  At its most basic, a map
for every second of the program that indicates how the user watched it,
namely:
        Watched at normal playback speed (or slower)
        Watched at a timestretch acceleration > 1.4
        Watched at a slow FF speed
        Watched at a fast FF speed
        Skipped using seek button
        Skipped using commercial skip button
        Didn't watch at all (ie. terminated viewing before watching)

Anyway, as the user watches, figure this out and store it in the
database.   Upon quitting the program, compress the data (since
most seconds will look just like the adjacent seconds) into a series
of transitions between watching modes.   Send that, along with programid,
channel id and ntp calculated time index for the start of the show.
(Users on dial-ups and otherwise not running ntp would not be permitted
to provide data, though with extra work you could possibly bring them in.)

The central system receives the data anonymously, ideally via tor.

For things like sports, you would want to send up data before the program
is finished, for the eager viewers.

It then overlays the maps on top of one another.  Soon the transition
points converge.  You throw out the extremes and you have a very accurate
map.   The map contains correct cut points, and also figures about how
many viewers did not watch a section at regular playback speed.

Thus, for example, an interesting ad that caught people's eyes might
have a lower skip percentage.   In a sporting event you would start to
identify non-breaks and put up a note on the screen, "20% of users made
a 15 second skip here" -- over things like a pick-off duel or a mound
conference etc.

Then the server allows boxes to pull down this accurate map when getting
ready to watch a program, and update it during the watching.

Expect a lot of bandwidth.