Some tuning questions

To discuss the tuning of comskip and for posting your ini files
Post Reply
ed6612
Posts: 2
Joined: Mon May 15, 2006 6:46 am

Some tuning questions

Post by ed6612 »

Hello,

I recently downloaded comskip and am trying to tune it for better detection accuracy for certain channels (I use Time Warrner analog cable in Los Angeles). I read the tuning guide which has been very helpful in my task but I do have some questions:

In the final block list in the log file I noticed that one letter abreviations are used in the cut column (eg. 'b' - blackframe, 'u' - uniform, etc)
What do the following letters mean, they show up in my logs and are not documented:

L, C, E, S (capital), A (capital)

I am also trying to figure out what some of the heuristic parameters mean, in particular:

length_strict_modifier
length_nonstrict_modifier
schange_threshold (I know it is scene changes but how does this parameter relate to min_schange_modifier and max_schange_modifier?)

I also noticed in my testing that uniformity is a very accurate indicator of where comercials exist, is there a heuristic parameter that would take uniformity into score consideration

I also set logo threshold at 0.8 and ran comskip on a 30 min show that has a small logo in the corner while the show is running and no logo during commercials. comskip successfully detected the show logo at 0.85 but on the very last commercial flagged about 20 seconds of the commercial as having a logo as well at 0.42. I played back the video and confirmed there was no visual logo during the commercial, just some constant text at the bottom of the screen.
Should the logo at 0.42 been disregarded by the threshold or do I need to run a second pass of comskip for mask matching to invalidate the commercial 'logo'?

Anyway, too many questions, I will stop now, if anyone has any thoughts please feel free to share.

Great program, very accurate for the most part :D

-Ed
erik
Site Admin
Posts: 3368
Joined: Sun Aug 21, 2005 3:49 pm

Post by erik »

Many questions :)

The capital letters stand for the rule that is used for scoring
The small letters stand for the audio/video feature used to decide this is a possible cut point at the end of the block.

If uniform is a good hint, look at the "punish" scoring. enable uniform in punish and set the amount to punish.

The modifiers modify the score when the rule aplies.
Strict means strict commercial size, non strict mean non strict but still commercial size

schange_threshold the the amount of change required to trigger the detection of a scene change, do not change this value.

I don't fully understand your logo question, maybe you can upload this recording to my ftp server so I can have a look.
ed6612
Posts: 2
Joined: Mon May 15, 2006 6:46 am

Post by ed6612 »

Thank you Erik for the answers,

Just to make sure I understand the scoring method abreviations correctly,

A = aspect ratio, C = closed captions, S = scene changes, L = length, E = ???

As to the logo, basically a false logo was detected in a commercial that lasted about 20 seconds and the debug window listed it as having a value of 0.42. In the configuration I had the logo threshold set at 0.8. I was just wondering if that detected logo was a factor in scoring if it's value was below the threshold.
Unfortunatelly the recording associated with this question is a bit large (1 gb) but I will play with the settings some more on a short segment and let you know if I can reproduce the issue.

Thanks again,

-Ed
erik
Site Admin
Posts: 3368
Joined: Sun Aug 21, 2005 3:49 pm

Post by erik »

The logo bar in the timeline shows how the logo was used for scoring.
If during that commercial there is no black logo bar in the timeline of the debugwindow than the logo has been below the threshold.

The letters are different

A = aspect ratio different from main AR
C = Combined commercial size
S = Strict commercial size
L = Logo presense/absense used
E = Exeeds min show segment length
N = Non strict commercial size

The numbers signal one of the global heuristics rules that can override any other scoring.

Can you mail me the .csv file of that show and a .ref file showing where you would like comskip to have found the commercials?
yankinoz
Posts: 10
Joined: Mon Jul 17, 2006 12:08 pm
Location: Sydney Australia

Post by yankinoz »

Hi Erik
Can you tell us what numbers correlate to which global heuristic?

Thanks
Rob
erik
Site Admin
Posts: 3368
Joined: Sun Aug 21, 2005 3:49 pm

Post by erik »

A long overdue answer

The meaning of the numbers of the heuristics scoring

1 = Added block because short and after strict commercial
1 = Discarding block because too short and between two commercial blocks
2= Discarding block x because block y has also logo and small non logo gap.
2 = Discarding block x because too short and before commercial.
3 = Discarding block x because block y has no logo and others do.
4 = Added block x because of large black gap with block y
5 = Deleting block x because it comes after the last commercial.
5 = Deleting block x because it comes before the first commercial.
5 = Deleting block x because its too short and before first commercial.
6 = Deleted commercial because it is too long
6 = Deleted commercial because it is too short

I now see I have made some double use of numbers.
Maybe some day I will correct this ;-)
Post Reply