DVRMSToolbox modifies Comskip xml output?

[Ken62465]

OK it's driving me nuts and not sure what I'm seeing. I've been trying to tweak setting in comskip.ini for better logo detection...cut points seem very inconsistent. Found a few settings seeming to help ok but then I wanted to use the comskipGUI to see exactly what's going on. I dropped a .wtv file onto the GUI and could not help but notice the xml file created is different than the xml file created if I run it through DVRMStoMPEGGUI. The cut point I'm struggling with is right in the xml version created from comskipGUI but not run through DVRMStoMPEGGUI. Why on earth is it modifying points? Even the files formating is different. The comskip xml is one commercial start/end entry per line. The DVRMStoMPEGGUI is all on one line. Anyone else see this? My logic was the DVRMStoMPEGGUI is just another means to run a rule set that the DTB filewatcher does itself so should be the same and a decent means to test files and duplicate the process that will happen on it's own. Maybe it's me I don't know. ohh...even the frame indicators in the .txt file do not match the xml timeline cut point....I did a little interpolation to compare that as well. It's not comskip mucking with the files in an odd fashion imo.

[Ken62465]

ok further testing and it's not DVRMSToolbox mucking with the files. I almost thought I found it where comskip writes the xml file but in the DTB profile used the last function is "Write commercial spans as xml". I thought this was it and rewritting the files so I removed the entry and retested...no change at all I could find. I was still seeing the file be different and then I found it. First I was was opening the .xml file and .txt file while the comskip write process to these files was going on and the expected timeline of commercial start and commercial end was right. Once the .wtv was complete it seems to be the comskip scoring process that then rewrites, and fomats, the file and yields slightly different results. The non scored values were closer to my manual timeline I did to test this whole thing so I had expected numbers. Now here's the odd part. If I drop the .wtv file on the comskipGUI I get identical results/happenings with the altered start/ends. However if I run it again and leave just the logo.txt file for the second run my results are different and now closer to my manual timeline as if the scoring process was now skipped. Why if the logo.txt file exists do I get different results? I just ran this theory twice and it did the same thing. The results when the logo.txt file existing is near perfect cut points!

What's this telling me?

[Ken62465]

I saved the two different .log files to compare. This first one below is when the logo.txt file exists it does a better job detecting logos, even if in a small segment like 21seconds as seen below. Isn't the first phase of comskips detection to find the logo? if so why would it matter if the logo file exists or not? Other than being able to skip a detection process and speed things up I see no reason why they'd yield different results. I just ran this process on another .wtv file having issues with logo cut/no cut points and when the logo.txt file exists it again produced different, and better results.


Logos detected on the following frames
--------------------------------------
Logo start - 178 end - 808 length - 0:00:21.02 before:5.9 s after:6.2 s
Logo start - 10154 end - 10987 length - 0:00:27.79 before:31.8 s after:141.7 s
Logo start - 12097 end - 22442 length - 0:05:45.17 before:96.6 s after:6.4 s
Logo start - 28917 end - 29402 length - 0:00:16.18 before:31.9 s after:254.8 s
Logo start - 30396 end - 36855 length - 0:03:35.51 before:81.2 s after:6.1 s
Logo start - 46462 end - 46947 length - 0:00:16.18 before:13.4 s after:260.2 s
Logo start - 47999 end - 54777 length - 0:03:46.15 before:64.7 s after:5.7 s
Logo start - 60440 end - 63854 length - 0:01:53.91 before:31.5 s after:390.9 s
Logo start - 64645 end - 75425 length - 0:05:59.69 before:171.8 s after:4.8 s
Logo start - 77608 end - 77629 length - 0:00:00.70 before:11.9 s after:6.9 s
Logo start - 78652 end - 79784 length - 0:00:37.77 before:27.2 s after:5.9 s

-------------------------------------------------------------

Logos detected on the following frames
--------------------------------------
Logo start - 10154 end - 10987 length - 0:00:27.79 before:31.8 s after:141.7 s
Logo start - 12097 end - 22442 length - 0:05:45.17 before:96.6 s after:6.4 s
Logo start - 28917 end - 29402 length - 0:00:16.18 before:31.8 s after:254.8 s
Logo start - 30396 end - 36855 length - 0:03:35.51 before:81.2 s after:6.1 s
Logo start - 46462 end - 46947 length - 0:00:16.18 before:13.4 s after:260.2 s
Logo start - 47999 end - 54777 length - 0:03:46.15 before:64.7 s after:5.7 s
Logo start - 60440 end - 63854 length - 0:01:53.91 before:31.5 s after:390.9 s
Logo start - 64645 end - 75425 length - 0:05:59.69 before:171.8 s after:4.8 s
Logo start - 77608 end - 77629 length - 0:00:00.70 before:11.9 s after:7.0 s
Logo start - 78652 end - 79785 length - 0:00:37.80 before:27.1 s after:4.7 s

[erik]

Is
two_pass_logo
set to 0 or 1 in comskip.ini?

[Ken62465]

lol..checked that right off the bat. it's not in my ini file but checked the log output and it's default to 1 in the record. I did forget to mention that if I enable live_tv=1 is where I can open the xml file at stages as it's writing to it and see what happens...lol because it's live. This was how I spotted the fact that the break was initially right but the last stage altered it to an incorrect value...that's in the scoring right?

I've now tested 7 different files doing this with and without a logo.txt file present and 4 file did have the different values I mentioned. ...3 of the files yielded no difference when logo.txt file existed or not. There's something about the video triggering this and to me this is looking like a bug. Strange thing is the commercial breaks are corect before scoring even when log.txt file did not exist, then altered to wrong values. If logo files existed the values did not get altered.

Maybe what I'll do is zip this issue up per your "How to ask for help" and see what you might find causing this behavior. I think I'm real close to some good results but this issue is a nagging one effecting 1/2 my files tested.

Using latest donator version.

Not sure this will mean anything but I watched the file execute in the comskipGUI and the first 35 seconds of the video is show with logo. The logo find phase never picks up this logo when looking for one, it's finding the logo almost 10 minutes in. Then when the second pass starts that first 35 seconds of video does have the highlighted box around the logo so it sees it. The point the scoring seems to decide to cut it is right when the two people stop talking and there is only background sound effects....the sound effects are pretty loud with a slight increase that I wonder if it's fooling comskip into thinking a commercial break is coming? I also then wondered about all these -1 values I see for sound levels in the log file but found a past thread topic talking about that and not to worry if the max volume had a decent value...it did. Under 5.1 channel HD recording what channel does comskip check for sound? Does it check all channels?

[erik]

With some recordings the sound runs a number of frames ahead of the video so the first frames of the video may have sound level -1 as to associated sound is available in the recording. Don;t worry about it.

[Ken62465]

ok so we'll rule out a sound issue. What do you think is doing this odd behavior with the scoring? Seems the videos that do not get cut right are fine if I do a second pass and leave the logo.txt file for it to use. Why would the existence of a log.txt file change the results slightly? I did find a typical trait between all the videos that exhibit this behavior. These are the recordings I mentioned before where my HTPC clock was just under 5minutes ahead of time. So the default recording in WMC7 of 5 minutes before show start caused my recordings to star slightly into the show. I'd still think this should not be happening an cutting out the first show segment during scoring....at least that's what it looks like it's doing when I watch the live written xml file.

[erik]

Please follow the "how to ask for help" post

[Ken62465]

zip sent