[mythtv] Commercial Detector questions

f-myth-users at media.mit.edu f-myth-users at media.mit.edu
Fri Dec 9 19:57:15 EST 2005

    Date: Sat, 10 Dec 2005 01:27:11 +0100
    From: "Pascal Favre" <Pascal.Favre at gmx.net>

    I use commflagging for all my channels, the results depend highly on the channel.
    There are several options which are global. Would it make sense a user can set the options on each channel, which are provided as a 
    parameter to mythcommflag. A bit-wise encoded value would do.
    To receive the best results, the user could run mythcommflag in verbose mode, where the commercial detection types found are 
    protocolized. The user can check this and optimize the settings.
    Further, there could be a Wiki-like database for those commflag values based on region and channel.

I think this is a -great- idea (and might not even be all that much
work to implement, which is good, since we're talking about putting
more on Chris's plate in a sort of unfunded mandate. :).

The wiki-like database would be a sort of wiki-human-intermediated
version of collaborative filtering (this is the technology that later
got turned into the if-you-liked-that-you'll-like-this stuff at Amazon
and many other places).  And anybody could set that up; doesn't take
anyone particularly familiar with Myth code or anything like that.

Also, by breaking this stuff out, it would enable some enterprising
soul to use some machine-learning techniques; this is a classic case
where some supervised learning might yield some good algorithms.
(Basically, you show the user either the entire recording or just
the cutpoints and the user says "right" or "wrong" for each decision,
as supervisory input to one of any number of machine-learning
algorithms---the learner looks at the various inputs from the
commflagger and from the user and builds a classifier.)  Might make
good fodder for an advanced undergraduate or master's thesis or
something like that---but only if the hooks were already there.
Otherwise the scope & prerequisite familiarity with the code (and
the risk of timeconsuming failure) might be too large for a typical
smallish thesis project.

More information about the mythtv-dev mailing list