Comskip vs ShowAnalyzer?

[jlindborg]

Hmmm... I'll record a couple of those shows with my setup and see how it works - I wonder if the recording quality has anything to do with it. Both SA and Comskip work about the same on most shows with just a few exceptions for me other than HD which SA doesn't deal with well. I'm using DirectTV as well with 2 250s and the quality cranked up. It's suprising that you guys see such wildly different results, but such is life I guess.

[jlindborg]

Well, I recorded a rerun of King of Queens on KSTW (local Seattle channel) and the detection was perfect - pretty much frame on - couldn't ask for better.

I don't see any episodes of Smallvile - I must not get that out here or it's on a channel I don't get.

I guess the CBS broadcast of King of Queens might be different, but all the other CBS shows I watch are spot on so I kind of doubt that. It must be related to the video quality based on recording level and/or capture cards or something assuming our INI files for Comskip aren't different and we're running the same version and all other factors are equal. Or maybe I just got real lucky with the episode I happened to get today.

But spot checking all the other shows from the last few days, I'm seeing very few mistakes and none of them nearly as massive as what you describe. Now and again it'll leave a commercial out of a block (usually at the end of a block) - but not often.

Pretty puzzling that you guys have such catastrophic results and mine are so good. I must just be "in the zone" out here. Probably all that clean air and rain we have up this way...

[dblaine2]

I really do think it has to do with quality. Not sure if its related to recording quality or signal quality but quality nonetheless.

Some of my channels are a bit grainey and those do not detect commercials very well. I haven't done any investigation to figure out why.

Just for giggles, what is your recording quality?

[jlindborg]

I have it cranked to DVD long play (about 2.4 gig an hour). I get locals (including 11 which I recorded this on) over the dish and they're pretty clean.

[malbec]

I was so pumped to try comskip thinking it would do just a bit better. Unfortunately, I must report it did not turn out so well for me. Comskip took longer to analyze the files than SA and, when completed, it only captured a few minor things at the beginning and end of the recordings with nothing caught between as others have reported.

Test files...

The Rookie
A few Smallville episodes
A few Stargate episodes

Ah well, the search for perfection continues. Thanks again jlindborg for all of the work you did on this!

[blade]

All of my recordings are done at 3.2 gig/hr. The 2 shows I tried it on were very clear. It wasn't due to having a poor quality capture. I've seen the same thing happen when people run SA. Many people complain about Lost and 24, both of which are 2 of the most accurate for me.

Stations differ in how they fade in and out of commercials and the commercial lengths vary. Not all stations broadcast with the same contrast and brightness levels either. For example I have 2 local NBC stations one of which is much darker than the other, even when watching the exact same show. I would imagine when detecting black frames just how black they are and how much they differ from the rest of the scenes makes a difference.

[jlindborg]

Looks like most of the logic is around logo detection these days - the black frame stuff is used for cleanup but the major detection work circles around finding and tracking those fab logos. I'm assuming it's having a hard time with your recordings for whatever reason - when you run it in debug mode does it "lock in" on the logo? Or declare it can't find it?

Guess all that clean living is paying off in commercial karma... to be clear, I'm anxiously waiting for Jere to update SA for HD support so I can go back to it - the ability to skip recording on some channels and get it to back off processing during records is very nice stuff.

[erik]

I just now discovered this thread.
For those with bad detection results using comskip I welcome them to upload a recording to my ftp server so I can have a look why.
To have a serious try I do need at least 30 minutes with at least one commercial block so its about serious uploading.
The ftp server has 20GByte free space so you are all invited.

[blade]

Originally Posted by jlindborg

Looks like most of the logic is around logo detection these days - the black frame stuff is used for cleanup but the major detection work circles around finding and tracking those fab logos. I'm assuming it's having a hard time with your recordings for whatever reason - when you run it in debug mode does it "lock in" on the logo? Or declare it can't find it?

If it's relying too heavily on the logo then that would explain a lot because some of my recordings don't have the logo at all.

I just now discovered this thread.
For those with bad detection results using comskip I welcome them to upload a recording to my ftp server so I can have a look why.
To have a serious try I do need at least 30 minutes with at least one commercial block so its about serious uploading.
The ftp server has 20GByte free space so you are all invited.

I've already deleted those recordings. If I can find time to run comskip on a few more shows I'll try to upload them. Unfortunately I'm in the middle of finals and don't have much time to play around at the moment.

[dshields]

Originally Posted by dbfresh23

After seeing this thread I decided to give comskip a shot (Haven't used it in a LONG time). Now I remember why I like ShowAnalyzer so much! Comskip did a horrible job on all 3 shows I ran it against. It skipped over the first 2-3 minutes and the last 2-3 minutes and nothing in between on all 3 recordings. I switched back to SA and while it wasn't perfect it was much closer then comskip. I used the default settings for both apps through DirMon2.

This is exactly my experience with Comskip. I haven't tried SA. Sorry I don't really have the expertise or time to edit .ref files for missed commercials. Short of that, is there anything I can do to get Comskip to work? If I had a good explanation of the .ini settings, I'd be willing to experiment.

[erik]

Because its assumed to be such a obvious failure I think I can do without the .ref file for the moment
So sending me the .csv generated when you set output_framearray=1 in the ini file should be enough
If possible you can include in the mail the start and stop frame numbers or time of the commercials.

[CadErik]

Originally Posted by dshields

This is exactly my experience with Comskip. I haven't tried SA. Sorry I don't really have the expertise or time to edit .ref files for missed commercials. Short of that, is there anything I can do to get Comskip to work? If I had a good explanation of the .ini settings, I'd be willing to experiment.

Actually I had a good experience with both software. That far I noticed that SA tends to skip too much in the beginning of shows and comskip did a perfect job on several shows I tried. Just curious, are both of them logo detection based?

Erik.

[erik]

Comskip uses many clues to discriminate between show and commercial.

First there is segmentation. This is done based on black frames, scene changes, sound level, aspect ratio and also logo presence.
Then there is classification of the segments. This is done on many parameters amonst them brightness, volume, length, logo presense, scene change rate, silences, aspect ratio, and others.
And at last there are the heuristics rules.

How to tune comskip is explained in the tuning manual you can find at my web site and inside all comskip distribution zip files.

http://www.kaashoek.com/comskip/files
Look for tuning.htm

My ultimate goal is to eliminate the need for tuning. But when I see what the Chinese are broadcasting I am not sure I will be able to reach that goal.

[blade]

My problem was that comskip couldn't find the .ini file. I dropped it into my system32 folder as suggested in another thread and ran it on an episode of Futurama and it was more accurate than SA. Not quite perfect, but close. I'm going to run it on a few more shows and see how it does.

For some reason when using comskip with dirmon2 it can't find the .ini file. I have a comskip folder that contains all of the comskip files, but it can't find it in that directory.

I think erik said in another thread that it can find it if it's in the same folder as the recordings, but that's far too messy IMO. So I'll just leave it in the system32 directory.

[dblaine2]

Originally Posted by blade

My problem was that comskip couldn't find the .ini file. I dropped it into my system32 folder as suggested in another thread and ran it on an episode of Futurama and it was more accurate than SA. Not quite perfect, but close. I'm going to run it on a few more shows and see how it does.

How did it go?

[lovingHDTV]

Originally Posted by blade

My problem was that comskip couldn't find the .ini file. I dropped it into my system32 folder as suggested in another thread and ran it on an episode of Futurama and it was more accurate than SA. Not quite perfect, but close. I'm going to run it on a few more shows and see how it does.

For some reason when using comskip with dirmon2 it can't find the .ini file. I have a comskip folder that contains all of the comskip files, but it can't find it in that directory.

I think erik said in another thread that it can find it if it's in the same folder as the recordings, but that's far too messy IMO. So I'll just leave it in the system32 directory.

This is the exact problem I had with poor comskip results. I back rev'd Dirmon to Dirmon2 and the command line option started working again and I got good results. Now I can skip commercials on HDTV streams

[blade]

Originally Posted by dblaine2

How did it go?

It was very accurate. I think I've decided to switch from SA back to comskip. SA used to be a lot faster, but now comskip is processing files at almost the same speed. SA and comskip appear to be pretty even in accuracy on completed recordings; however, I like to process shows as they record in case I want to watch something that's still recording. Comskip appears to be quite a bit more accurate when it comes to this.

To be fair though, I have Dirmon2 set to launch up to 4 instances of SA at a time. So as many as 4 shows may be processing as they're recording. I've now set dirmon2 to do the same for comskip. So far I've only ran 1 instance of comskip at a time on shows in progress. I'm curious to see how well comskip works when there are 3-4 shows being processed.

[geogecko]

So, depth of analysis setting lower, i.e., 10, provides better detection? Just wondering which way to crank the setting.

Thanks.

[dshields]

Originally Posted by blade

My problem was that comskip couldn't find the .ini file. I dropped it into my system32 folder as suggested in another thread and ran it on an episode of Futurama and it was more accurate than SA. Not quite perfect, but close. I'm going to run it on a few more shows and see how it does.

For some reason when using comskip with dirmon2 it can't find the .ini file. I have a comskip folder that contains all of the comskip files, but it can't find it in that directory.

I think erik said in another thread that it can find it if it's in the same folder as the recordings, but that's far too messy IMO. So I'll just leave it in the system32 directory.

I did this also and it fixed all my problems. Comskip commercial detection is really good now.

[Jackal24]

What are the advantages/disadvantages of each? I am currently using comskip, but wondering if it is worth it to switch. I don't need a preview window or anything because I always run Sage full screen (dedicated comp). Does one have better recognition than the other?

* merged *