[mythtv-users] mythcommflag logo detection does not work

[Karl Newman]

On Mon, Dec 23, 2013 at 7:34 PM, Michael T. Dean <mtdean at thirdcontact.com>wrote:

> On 12/23/2013 06:41 PM, Stefan Pappalardo wrote:
>
>> Am 23.12.2013 14:57, schrieb Stefan Pappalardo:
>>
>>> I'm using mythcommflag version: fixes/0.26 [v0.26.1-28-g3beccab] on
>>> mythbuntu 12.04. In german tv there is no logo in commercial breaks.
>>> That's why logo detection should flag up to 99 % of the commercials.
>>>
>>> But it detects nothing, and no differenc turning on/off strict.
>>>
>>> Could there be something wrong with the used version of mythcommflag?
>>>
>>> Is there anyone with german DVB-S(2) with successful logo detection on
>>> ProSieben or Sat.1?
>>>
>> Does anyone find commercials with logo detect method?
>>
>
> The best I can tell you is that commercial detection worked extremely well
> in the US when it was developed (back in the days of SDTV) because it was
> developed by a developer from the US.  It has progressively gotten less
> reliable in the US because of changes to how broadcasters send out the
> video--with different broadcasters doing things differently resulting in
> varied success across different channels.
>
> Outside the US, no one has put in much effort (beyond a couple of started,
> but not completed attempts) to improve results of commercial flagging, and
> from what I understand, few non-US users have much success with it.
>  However, for me, whether it works or not, I can generally skip over any
> commercial break in just a few seconds with only a couple of button presses.
>
> Mike
>

Something changed in myth which made it much less reliable (my money is on
the ffmpeg libraries). I've been recording SD video in the US since I first
started using MythTV (8 years now) and commercial detection was darn near
perfect in 0.24. I almost never had to pick up the remote while watching a
show. Upgrading to 0.25 broke commercial detection for me (it catches maybe
half now and has more false positives too), and it hasn't really recovered
since. I've played with the settings re-running comm flagging on a few
representative recordings and at the time I found that the Blank Frame +
Scene Change (experimental de-selected) seemed to find the most. Logo
Detection was near useless and including it in the mix (selecting the "All"
method) actually seemed to make the overall detection worse.

I thought someone was gathering samples (maybe 2 years ago?) for an update
to the comm flagging. What happened to that effort?

Karl
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.mythtv.org/pipermail/mythtv-users/attachments/20131223/5fef4f98/attachment.html>