Logos and 'LOST'

[Stuntman]

Last nights lost was different in the way the logo was or wasn't displayed.. when I ran Comskip I noticed that the first pass seemed to be right on, but before comskip finished it seemed to go back and on that second (?) pass it broke one of the early detections.. So I went back and watched the show to see where it 'lost' it.

Funny thing is that there WAS an AR change as the show came back from commercial, but something I don't usually see happenend.. ABC did NOT put their channel logo 'abc HD' into the video for probably 2 minutes into that segment! Comskip didn't catch that the show had returned from commercial until the seemingly random reappearance of the 'abc HD' logo later in that segment! Very strange.

This would bolster the argument for Comskip to look for the RATING logo that DID pop up at the beginning of that segment.. If COMSKIP had seen that, it would NOT have missed the beginning of that segment, even though the station logo was not there.

Comskip caught the rest of the commercials correctly as it appears the station logo was present at the more normal beginning of each segment.

FYI

[erik]

I'd like to take a look

Could you follow this procedure?

Set
output_framearray=1
and
verbose=10
in your comskip.ini
Run Comskip on the recording.
copy the generated .txt file to a file with the same name but with .ref as an extension.
Edit the .ref file such that it contains the start and stop frame numbers of the commercials as you would like them to be found.
Zip the comskip.ini, the .csv, .log and .ref files together and mail me the zip file so I can have a look.

[erik]

And the new build of Comskip can decode the XDS data that contains the V-Chip information
Could you set
output_framearray=1
and
output_data=1
and enable the "close caption" method (16)
and upload the generated .csv and .data files?

[Stuntman]

And the new build of Comskip can decode the XDS data that contains the V-Chip information
Could you set
output_framearray=1
and
output_data=1
and enable the "close caption" method (16)
and upload the generated .csv and .data files?

Will do! Can you PM me where you want them sent?

PS.. I don't get a .txt file when I do this.. can I create the .ref from an .edl file? I assume I have to do this with the gui version??

[erik]

To get a .txt file you have to set
output_default=1
When you already have a .csv and .data file you can drag and drop the .csv file onto comskipGUI.exe and then the .txt file will be created.
You can use the 'T' key in the GUI to toggle blocks from show to commercial and vice versa to make sure the generated .txt file is correct. Then rename the .txt to .ref
Hit 'W' to write the new .txt file after toggling certain blocks.

I just noticed a bug when using build 135 when no .data file is available. Will be solved in 136 out within some minutes.

Will PM you with the upload details.

[erik]

Stuntman

Thanks for the uploaded files.
They are very helpful to understand how CC and XDS data is broadcast.

W.r.t tuning you should NOT use silence based cutting on a show like Lost.
Silence base cutting is nearly never needed on USA broadcast

May I suggest you change
detect_method=255
to
detect_method=63

That should solve most of the problems.
Will look more into it later today

[Stuntman]

May I suggest you change
detect_method=255
to
detect_method=63

That should solve most of the problems.

I tried this and have to report that my detections got much worse! Survivor last night missed probably 2/3 of the of commercials in every break. Survivor was recorded from the HDHomerun, but is not an HD broadcast.. yet..

[erik]

Could you upload the .csv file of Survivor?
I see you have set
max_avg_brightness=25
and
intelligent_brightness=1

Is this for any particular reason?

In general a better setting is
max_avg_brightness=20
intelligent_brightness=0

The USA specific intelligent brightness is error prone when not used in combination with a very precise score_percentile and black_percentile setting.

Lost is always very difficult due to the many silences

[Stuntman]

Could you upload the .csv file of Survivor?
I see you have set
max_avg_brightness=25
and
intelligent_brightness=1

Is this for any particular reason?

I thought I'd read that those were better USA specific tunings.. I'll send you the .csv file, then make the adjustments you suggest as far as brightness and run it again.. see how it goes!

[erik]

If you can send me a set of .csv files (and the .data files that go with them) I will do the tuning for you as I really need a representative set of new .csv file.

[blackrat]

stuntman
have you tried
punish_no_logo=0
that's what i did

erik
have some .csv and .data files will XDS you want them
PM info for upload and how

[erik]

The problem with the "lost" recording of stuntman is that the first segment without logo directly after the commercial has the exact length of a commercial. As long as Silence cutting is enabled I do not know of any way to prevent this from happening.

[Stuntman]

I do not know of any way to prevent this from happening.

I think this is part of the argument that some have raised about looking for the 'Rating' the almost always is present when a segment of show begins.. I have never seen that rating anywhere except during the show, so it would seem like a good way to better mark the start of a segment.. I've seen NBC also change their logo position during shows.. I'm seeing a tiny logo sometimes placed on the bottom left hand side during some shows, before moving back to the more normal bottom right side..

I suspect the networks are on to our commercial cutting!

[erik]

Indeed,

But the examples I received that showed problems all where caused by having silence detection enabled where it really was not useful.
I have some logic to disable silence detection when it produces bad results but at this moment its insufficient.
Maybe I can make the heuristics more strict.

I really like to receive some examples (.csv, .ref and .log) of USA recordings where not having silence detection enabled is causing problems so I can make this adaptive

[erik]

Blackrat,

I looked at the excellent samples you uploaded.
Also here, why do you use
detect_method=127
i.s.o
detect_method=63
Silence detection only makes things worse in these examples

The XDS data is not adding much value except for one recording where it nicely indicates where the start and end of the recording is. The other recording did not have XDS data.
The V-Chip data is constant during the whole show. Also during the commercials.