Comskip Support Forum

Comskip is a free commercial detector, browse the forum for more information
It is currently Wed Oct 17, 2018 6:54 am

All times are UTC [ DST ]




Post new topic Reply to topic  [ 2 posts ] 
Author Message
PostPosted: Tue Dec 11, 2012 2:09 am 
Offline

Joined: Wed Oct 19, 2011 5:37 am
Posts: 22
hi erik,

While trying to improve my comskip tuning I've come across what looks like an issue with the averages for each block. Here's a screenshot of the graphs showing 4 blocks.

Image

Looking at the brightness at the bottom to begin with, the first two white lines look roughly correct, but the next green line is way too high compared to the red values behind it.

The values from the log match the lines drawn.
Code:
  5:--    1   1   0  30357  32108   1.50   70.040  0.19   0 1.26              N      b  1453+ 0.08 1265+ 95-  0.000   145           
  6:--    2   1   1  32109  33109   3.00   40.000  0.14   0 1.26            L N      b  1617+ 0.00 1054- 95-  0.000   161           
  7:++    3   1   2  33110  44913   0.01  472.120  0.32   0 1.26           EL        b  1966+ 0.90 1204+ 980  0.000   196           
  8:++    3   2   1  44914  48414   1.00  140.000  0.13   0 1.26                     b   975- 0.11  940- 94-  0.000    97           
1453, 1617, 1966 and 975. But the brightness values for each frame are between 15 and 224 so these averages are all way too high.

I loaded up the csv file in excel and calculated the average brightness for each block and came up with 109, 136, 71, 102 which looks like what I would have expected.

Graphing these values looks to give a far better indication of the brightness of each block.
Image



Out of curiousity I had a look in the source code and noticed this
Code:
//            b = frame[k].brightness;
                    b = abs(frame[k].brightness - frame[k-1].brightness);
It looks to me like the average brightness is being calculated using a difference between frames rather than the value from each frame, which I imagine gives us something more like a scene_change value. Given that the correct code is there, but commented out, I'm assuming at some point you changed the way average brightness is calculated, however the current behaviour results in "averages" which result in bad scoring for my files.

Can the average brightness calculation be changed back to be a true average?



Moving on to the other graphs. The Uniformity and Scene_change averages appear to have the same problem as brightness. I graphed the averages in excel and it showed that the 3rd block should have lower values than the 1st 2nd and 4th, but as you can see in the screenshot above comskip has the 3rd block marked as the highest value and punishes them accordingly.

The Sound graph appears to be showing the averages all at or close to 0. The values in the log file block list are correct and match what I calculated in excel so I think this might be just a drawing issue.

Are these issues you can fix?

cheers,
nate


Top
 Profile  
Reply with quote  
PostPosted: Wed Dec 12, 2012 8:59 am 
Offline
Site Admin

Joined: Sun Aug 21, 2005 3:49 pm
Posts: 3211
I tried both (average brightness and average brightness change) and the later was most powerful in detection.
Of course I understand your confusion and I should make them all available as separate options.


Top
 Profile  
Reply with quote  
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 2 posts ] 

All times are UTC [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum

Search for:
Jump to:  
cron
Powered by phpBB® Forum Software © phpBB Group