Commercial Advance Beta Release

[JasonJoel]

I'm not sure if it only crashes on that one file or not. I didn't try any other ones after I pinned down that one as a bad actor.

I am running the debug comskip on the file now. Should have the results in ~15 minutes. I'll add it to this post.

EDIT:
Ok, that isn't fair. With this version it completed successfully and did not crash. I'm not sure if it will help you, but here is the LOG file anyway.

Jason

[Jere_Jones]

That's wierd. All I did was change all the exit points to have unique numbers and add that exit code to the log file.

If it happens again, let me know.

Jere

[ellweber]

Jere,

Same result here. Ran with debug version to completion but then I reinstalled the prior version, 0.52 and again it drops out with a null .txt file. I used the same .ini file as provided in the zip for all tests.

Mysterious stuff!

Lynn

[Snuffy2]

Originally Posted by Jere_Jones

I certainly appreciate the kind words. And at the very least, I appreciate my work. As far as sophisticated insights, I don't think so. I just know what I have learned from trying to improve comskip. Still, thanks.

Unfortunately, I don't have a clue. The decoding just seems to stop. Very wierd. But, I'm attaching a special version that will give me very specific information about the exit that should help me. Run this one and send me the log.

Jere

I'm having VERY similar crashes with almost identical log files, I'll run the 'debug' version & report back.

[Jere_Jones]

Umm... Ok... I'm glad it's working. I'm going to leave those changes in there since they seemed to do something. {shrugs}

Sometimes, (such as times like this) I think I fix more bugs by accident than on purpose.

Jere

[Deadbolt]

Actually, I hope you don't keep all these changes...my comskip now takes over 200 MB of RAM! YIKES!

EDIT: ...and it just crashed! This is with the debug version...I'm going back to .43 it is the last stable version for me

[ajuhawk]

I too have been having problems with .50 and .52. 0.43 was my last stable version and it worked relatively well. Aside from the crashes with 0.52 I've also noticed a couple of items:

1) It appears to work if I manually run it from within Sage.
2) When it does work, it misses a whole lot more of the commercial then 0.43 did.

I'll do what I can to feed you info, so let me know if there's anything that would be helpful.

-AJ

[Jere_Jones]

ajuhawk, Deadbolt,

How about posting or emailing me a log from a debug crashing? I can't fix it without more information.

Jere

[ajuhawk]

It figures that as soon as I write something it starts working. As I type it's working on like the 10th show in a row - catching up with all the ones it missed after it crashed yesterday or today - so I'll let it go until it either crashes or finishes. Hopefully I'll get it replaced with the debug one tonight or tomorrow and I'll let that go until it crashes (or maybe it'll work perfectly) *shrugs*

The joys of debugging...

-AJ

[Jere_Jones]

Originally Posted by Deadbolt

Actually, I hope you don't keep all these changes...my comskip now takes over 200 MB of RAM! YIKES!

All that memory was being used for the frame array output, which wasn't necessay if you aren't outputing a frame array file. Long story short, I've changed the allocation so it is only used if you are outputing or inputing a frame array file. It's memory use is back down to normal.

Originally Posted by Deadbolt

EDIT: ...and it just crashed! This is with the debug version...I'm going back to .43 it is the last stable version for me

See my previous post.

Jere

[Jere_Jones]

Here's another debug version that will let me know if comskip is crashing during the decoding process or during the analyzing stage.

It will help me find the problem if y'all will give this one a shot and send me the log file.

Thanks for all the input,
Jere

[ellweber]

Jere,

Here's a new mystery. Using a segment of my "test show" and the latest debug version I am finding what looks like an overlapping block. Log file attached shows block 0 from frame 1 through 1551 and block 1 from frame 0 through 4220. Block 1 is flagged for a logo at the beginning but the commercials actually start at 1551. Any clues?

Lynn

[Snuffy2]

Originally Posted by Jere_Jones

Here's another debug version that will let me know if comskip is crashing during the decoding process or during the analyzing stage.

It will help me find the problem if y'all will give this one a shot and send me the log file.

Thanks for all the input,
Jere

Here's a crashed log file (comskip.ini also included) from the FIRST debug version. Second one is running now, we'll see if I get any more crashed logs from that one

[Jere_Jones]

Originally Posted by ellweber

Jere,

Here's a new mystery. Using a segment of my "test show" and the latest debug version I am finding what looks like an overlapping block. Log file attached shows block 0 from frame 1 through 1551 and block 1 from frame 0 through 4220. Block 1 is flagged for a logo at the beginning but the commercials actually start at 1551. Any clues?

Lynn

This one I have clues about.

The log shows that it detected a logo starting in frame 29 that lasted until frame 2233. It seems to have a good lock on the logo. So my question is: Are you sure that the commercials started on frame 1551? Looking at your show's histogram, I think you may have your contrast set to low. Black video is supposed to be 16 but the blackest frame in your recording is 36. The brightest frame also seems a little dark, but I don't know if there were any pure white frames in the show. Improving this range will improve your black frame detection. Looking at the log file, I'm going to guess that the commercials really started at frame 2495 since it found a black frame there but removed it because it and the 23 next brighter frames.

Just something to think about.

Thanks for pointing this out. It has been fixed for the next update.

Jere

[ellweber]

Jere,

The last frame with a logo is at 51.2 seconds in this clip, frame 1534 at 29.97 fps. The block between 1551 and 2495 is definitely a commercial (The Donald promoting VISA cards!) and there are 50 black frames, by my count, in the gap at ~1550-1600. I cannot see any sighn of a logo in the 1550-2495 block of frames. I use VideoRedo for analysis but have been doing the seconds to frame conversion by hand. Too bad the test clip is 213 MB or I would send it to you!

This reminds me of the question as to whether you can ignore really small groups of black frames. I have seen some cases where there are only one or two that really don't constitute a break but in one cause comskip saw a break in such a situation.

Regarding contrast, I have used DScaler's [oops, make that ffdshow] histogram function to set up my two PVR-250s and have been pleased with the display quality as it is. Should I look at adjusting the .ini file to make comskip more accepting of these levels? I have intelligent_brightness=1 set, max_brightness=180, test_brightness=60 and max_avg_brightness=45.

Lynn

[malore]

Originally Posted by ellweber

This reminds me of the question as to whether you can ignore really small groups of black frames. I have seen some cases where there are only one or two that really don't constitute a break but in one case comskip saw a break in such a situation.

It's tempting to wonder this after looking at a file that has nice long gaps of a second or more, but the gap length varies by channel. One show I looked at had a commercial break starting with only one black frame which, probably due to the limitation of mpeg-2 compression, wasn't even very dark.

[malore]

Originally Posted by Jere_Jones

On a future development side, I'd like to explain the csv files and the reasoning behind them.
Comskip takes a long time to scan a recording. 99.999% of this time is used in decoding the mpeg stream. I had the idea that if I could save all the relavent information to a text file then I could try different settings quickly.

It might be possible to automate this even further with the use of a matching file with the correct commercial breaks. Then after collecting a library of these data files, automated tested could be done using a scoring system based on how close the generated results were to the correct results.

[Crashless]

Originally Posted by malore

It might be possible to automate this even further with the use of a matching file with the correct commercial breaks. Then after collecting a library of these data files, automated tested could be done using a scoring system based on how close the generated results were to the correct results.

They're called genetic algorithms. I don't know how to write them, but they're all the rage right now.

Basically, two similar but different algorithms are created, one does better than the other and is saved, the less efficient one is deleted, and two new algorithms are created using the more efficient first-gen as a base.

I have no idea how to implement it here, but I like the idea a lot.

Here's a mailing list/resource:

http://www.aic.nrl.navy.mil/galist/

[malore]

I wasn't thinking quit a fancy as genetic algorithms. I was thinking more along the lines of looping through the various settings to see which setting works the best on average.

[Crashless]

I know, but it's fun to dream.