tweaking comskip results - is this do-able?

[tevbill]

I'm having pretty good results with comskip but I have a question on fine tuning the final result.
Let's say the the txt file that comskip generated has the following final list:

Final Commercial List
---------------------
1 - start: 1941 end: 3441 length: 50.05 seconds
2 - start: 35055 end: 43641 length: 4 minutes, 46.49 seconds
4 - start: 70005 end: 78134 length: 4 minutes, 31.24 seconds
5 - start: 100048 end: 108467 length: 4 minutes, 40.91 seconds
6 - start: 117364 end: 119279 length: 1 minute, 3.90 seconds
8 - start: 156496 end: 164619 length: 4 minutes, 31.04 seconds
9 - start: 184518 end: 192919 length: 4 minutes, 40.31 seconds
10 - start: 205585 end: 207688 length: 1 minute, 10.17 seconds

221123 Frames Processed
_____

Then I notice that #6 is not a commercial, its a false alarm.
Can I just delete that line and renumber the remaining line numbers without corrupting the results?

Also if I notice that a commercial break runs a few seconds into the show like #9 can I just edit the end:192919 to end:192819 or something similar?

thanks a lot

[tevbill]

Well, now I'm thinking that I should make the modifications on the 1kb file that comskip generated rather than the large file.
The thing is I have not had time to test my theory becasue I decided to run comskip from the command line on a movie that it had not been run on yet.

Of couse I ran it on "A Beautyful Mind" which is a 7gb mpg file and I've been watching it run in a dos window for about an hour now :eek:

[reboot]

Yes, you can simply edit the 1k .txt file generated.
It may be better to tune comskip though.
Adjust your time settings, so block 6 falls between the min and max commercial times, then adjust black levels so block 9 is cut properly.

[tevbill]

I see what you mean, the compskip.ini should be tuned so it corrects the settings for all future recordings.
Right now those parameters are like this (the default)

max_commercialbreak=330
min_commercialbreak=44
max_commercial_size=180
min_commercial_size=1
min_black_frames_for_break=1;

so since there are not many 1 minute 4 sec commercial breaks I should set it like:

min_commercialbreak=120

and I can only "up" the min_black_frames so that should be:

min_black_frames_for_break=3;

right?

thanks

btw - I've been reading about SA (show Analyzer) which seems like a pretty neat application particularly the schoolhouse GUI.

Should I just switch to rather than comskip?

[reboot]

I can't tell you to switch to SA, but I can tell you that the next release of gbpvr will make SA even more effective, because of the profiles. You can set up individual settings for each channel if you want, instead of trying to get one set (comskip.ini) for all channels.

I was referring to black level detection, not black frames break. If darker portions of the show are detected as black, comskip treats them as commercials. Adjust black levels until they're detected as show.
You are correct about the commercial segment length.

[tevbill]

I'll play around with the following default settings. I don't see anything that specifically targets black level detection but possibly the test_brightness or max_brightness could be lowered to accomplish this.

appreciate the pointers


max_brightness=180 ;frame not black if any pixels checked are greater than this (scale 0 to 255)

test_brightness=60 ;frame not pure black if any pixels checked are greater than this, will check average brightness (scale 0 to 255)

max_avg_brightness=45 ;maximum average brightness for a dim frame to be considered black (scale 0 to 255)

[erik]

if you are tuning these parameters
max_commercialbreak=330
min_commercialbreak=44
max_commercial_size=180
min_commercial_size=1
min_black_frames_for_break=1;
youd better set output_framearry=1 in your ini file and then do the tests on the generated .csv file
This makes comskip skip the mpeg decoding and its much faster.

comskip testfile.csv

If you are tuning these parameters
max_brightness=180 ;frame not black if any pixels checked are greater than this (scale 0 to 255)

test_brightness=60 ;frame not pure black if any pixels checked are greater than this, will check average brightness (scale 0 to 255)

max_avg_brightness=45 ;maximum average brightness for a dim frame to be considered black (scale 0 to 255)
you have to rerun comskip on the mpeg because they determine what will be written in the csv file.

You can do beautifull things with the .csv file. Just take a look at the attached image

[reboot]

What Erik's trying to say, is play with the numbers in those three settings. (I should have been more specific when referring to "black levels: I meant the brightness settings.
Most good sources can have a max_avg_brightness=20, max_brightness=60, and somewhere inbetween, the test_brightness. (note these are much lower than the defaults)
If your source has lots of "snow", then the defaults (or even higher) may be better.

[tevbill]

ok the following settings you posted, reboot seem like a reaonable basis to test from. Comcast in the US is a pretty good source, occasional snow but not to much.

max_brightness=60
test_brightness=40
max_avg_brightness=20

and run some shows through comskip with the output_framearry=1 set to obtain the .cvs and see what happens.

Erik, the image above is cool - Is that is the .cvs file in Excel? is there an customized Excel chart available to obtain that graph from the data in the .cvs? or did you obtain it by just displaying as a standard chart?

[erik]

Its actually a spreadsheet I use to do anaylsis.

It can be downloaded here
Please be very patient while downloading, its about 50 MByte!!!!

Howtouse.

Doubleclick on the .cvs file. This will open excel.
control-A to select the whole sheet. copy and past into cell A1 of the input_data sheet of the comskip.xls file and the rest is automatic.

You can only display the data of the first 32k frames.
There is a second sheet called zoom that gives a more detailed view.

The translation between the .csv file and the graph is done brute force on the analysis sheet.

Studying the analysis will surely help you to improve the tuning of the blackframe detection including the silence detection.
Also you can easily judge the quality of the logo detection.

Wrong brightness settings will also make the aspect ratio detection (minY/maxY) wrong

Have fun!