Comskip: Any Reason to not use all Detection Methods?

[LewE]

Is there any reason to not use all detection methods in Comskip (i.e. detect_method=255)? Most sample ini files I see use only some.

I tried using all when OCmskip missed tow big commercial breaks in a program when it scored tham at 1.00. I was going to try setting the threshold lower but decided to see if using more methods could better score the blocks. It did find the commercial breaks and I found that using all methods doesn't appear to take significantly longer to process a file and finds some commercial breeaks that fewer methods missed.

What, if any, are the downsides of using all possible methods?

[dennit]

Is there any reason to not use all detection methods in Comskip (i.e. detect_method=255)?

Use what works. I've got some channels where logo detection causes commercials to be missed. The commercials have something that looks like the logo for that channel. Comskip thinks anything with a logo must be program, not commercial, so I've turned off logo detection for that channel .

[LewE]

Use what works.

I am more concerned in Comskip incorectly identifying program blocks as commercials than the other way around. I can always SKIP ahead a minute at a time through any undetected commercials but wouldn't even know whether I missed something if program material was identified as commercial.

I've got some channels where logo detection causes commercials to be missed. The commercials have something that looks like the logo for that channel. Comskip thinks anything with a logo must be program, not commercial, so I've turned off logo detection for that channel .

Are you saying that the network plays its logo over the commercials? If so, I can see why logo detection would be useless in that case. I suspect that advertisers aren't too happy about that.

[erik]

yes, there is a downside to using all methods and that is false positives (show being detected as commercial)
The more complex methods are intended to be used when the more robust methods fail to identify the cutpoints.
Example is certain chines broadcast that does not have black frames.

Best is to start with a limited set (logo and black frame) and add when needed.

And before you ask why there are so many methods let me explain that all of them where created because they where needed for some broadcasting style.

[erik]

Are you saying that the network plays its logo over the commercials? If so, I can see why logo detection would be useless in that case. I suspect that advertisers aren't too happy about that.

Discovery channel does in Europe

[erik]

Use what works. I've got some channels where logo detection causes commercials to be missed. The commercials have something that looks like the logo for that channel. Comskip thinks anything with a logo must be program, not commercial, so I've turned off logo detection for that channel .

It may be better to change the logo present modifier to 1 and to keep logo detection enabled as the logo detection prevents wrong cutpoints during the show

[erik]

I tried using all when OCmskip missed tow big commercial breaks in a program when it scored tham at 1.00. I was going to try setting the threshold lower but decided to see if using more methods could better score the blocks. fewer methods missed.

Do not change the threshold as the current logic can not cope very well with a different threshold and many detection options.

[dennit]

I am more concerned in Comskip incorectly identifying program blocks as commercials than the other way around. I can always SKIP ahead a minute at a time through any undetected commercials but wouldn't even know whether I missed something if program material was identified as commercial.

I agree it's more important to avoid incorrectly marking programming. However, when it makes a mistake, I like to find out why, and if possible , tune it so that it's even better.

Are you saying that the network plays its logo over the commercials? If so, I can see why logo detection would be useless in that case. I suspect that advertisers aren't too happy about that.

No, the commercials had their own logo, but it was too similar to the station logo. In this case, the station added some detailed Christmas related material to their logo, and this was a standard def channel. It probably had trouble resolving the details of the real logo.

The full comskip processing saw two characteristics of a commercial (IIRC, they were standard commercial length and had CC commercial hits in the dictionary). These two characteristics pushed the score up from 1 to 3 (default threshold of 1.0+), but the logo hit pushed it down to 0.3.

Commercial detection was perfect without the logo detection turned on. I suspect I could tweak logo detection - I see lots of parameters - but I'm in the middle of lots of other GBPVR changes, so for the holidays, I just turned off logo detection for these channels. (I process comskip in parallelprocessing.bat which was already set up for individual channel processing where needed.)

[dennit]

It may be better to change the logo present modifier to 1 and to keep logo detection enabled as the logo detection prevents wrong cutpoints during the show

Thanks for the advice - I'll change that now.
I wasn't sure of the relationship between cutpoints and scoring. The cutpoints seemed great. Does detect_method control cutpoints and the parameters control how the cut blocks are weighted - or do they interact?

[erik]

I wasn't sure of the relationship between cutpoints and scoring. The cutpoints seemed great. Does detect_method control cutpoints and the parameters control how the cut blocks are weighted - or do they interact?

The detect_method controls the enabling/disabling of certain methods. Some are cutpoint related (like black frame), others do are more scoring related (like logo). But once the information is there (like with logo) is also used for other purposes.
A better word for detect_method would be analysis_method