[mythtv-users] How well does commercial detection work?

[Kevin Kuphal]

Chris Pinkham wrote:

>>>I wonder if it would help to apply a narrow filter to get rid of high
>>>frequency noise before testing whether a frame is blank?  This might
>>>help for noisy signals without hurting things much for clean
>>>signals...
>>> 
>>>
>>>      
>>>
>>I'd be interested in this.  In going through my programs last night, I 
>>found that all my non-CBS recordings were about as close to 100% as I 
>>    
>>
>
>The commercial flagger calls NuppelVideoPlayer::InitVideo which calls
>NuppelVideoPlayer::InitFilters, so if you have any playback filters
>enabled, I believe they are also used for commercial flagging.  You can
>test by setting up a playback filter to see if it makes any difference,
>if so, we could possibly make a different setting for flagging filters
>so we could apply filters for flagging but not for playback (if the user
>so desired).
>
>  
>
>>can imagine.  The only one with issue was CBS and with strict detection 
>>turned on I believe it missed on break and had one short 12 second false 
>>positive and another 0 second false positive (makes me think we should 
>>have some kind of minimum bound on commercial length inclusive with 
>>strict detection to eliminate such short and obvious non-commercials). 
>>    
>>
>
>There already is a minimum length, if you are saw a 12-second or 0-second
>false positive, there is a bug somewhere.  If you see something like
>this again, can you re-run mythcommflag manually on that show with the
>following command line (add in your own chanid/starttime of course) and
>gzip and email me the output:
>
>
>DEBUGCOMMFLAG=1 mythcommflag -v commflag --chanid CHANID --starttime STARTTIME > mythcommflag.out 2>&1
>
>That will turn on a bunch of debug statements that I have in the code
>and will print out information about every frame, as well as the blocks
>that were defined/found and how the scores were adjusted for the blocks
>based on the stats for the frames in each block.   This log may be huge,
>so you probably need to gzip or even better bzip2 the file before emailing.
>  
>
I send this to you directly.  From the relevant portions I can see it 
did detect the 12 second and ~0 second break again.  If I read it right, 
that's at frame 18625 and then at 67576.  I am using the ALL detection 
method with strict detection enabled.

2005-04-03 12:07:33.803 Final Commercial Break Map
2005-04-03 12:07:33.803 ---------------------------------------------------

2005-04-03 12:07:33.803       0 : 4 (00:00:00.00) (0)
2005-04-03 12:07:33.803    1166 : 5 (00:00:38.26) (38)
2005-04-03 12:07:33.803    6601 : 4 (00:03:40.01) (220)
2005-04-03 12:07:33.803   11894 : 5 (00:06:36.14) (396)
2005-04-03 12:07:33.803   18625 : 4 (00:10:20.25) (620)
2005-04-03 12:07:33.803   19014 : 5 (00:10:33.24) (633)
2005-04-03 12:07:33.803   28951 : 4 (00:16:05.01) (965)
2005-04-03 12:07:33.803   37411 : 5 (00:20:47.01) (1247)
2005-04-03 12:07:33.804   55782 : 4 (00:30:59.12) (1859)
2005-04-03 12:07:33.804   61748 : 5 (00:34:18.08) (2058)
2005-04-03 12:07:33.804   67576 : 4 (00:37:32.16) (2252)
2005-04-03 12:07:33.804   67586 : 5 (00:37:32.26) (2252)
2005-04-03 12:07:33.804   74797 : 4 (00:41:33.07) (2493)
2005-04-03 12:07:33.804   83172 : 5 (00:46:12.12) (2772)
2005-04-03 12:07:33.804  104496 : 4 (00:58:03.06) (3483)
2005-04-03 12:07:33.804  109512 : 5 (01:00:50.12) (3650)
2005-04-03 12:07:33.804 ---------------------------------------------------

Kevin