Page 1 of 1
Frames out of bounds in txt but ok in log
Posted: Tue Nov 05, 2013 12:07 pm
by IrmaIsfot
Sometimes (have yet to see a pattern?) the frames in the output txt-file goes waay off.
I think it only happens on a HD channel?
There seems to be an offset added to the frame numbers?
The result towards the end in the log file seems ok, though.
Example txt-file:
Code: Select all
FILE PROCESSING COMPLETE 76730 FRAMES AT 2500
-------------------
2431184 2432952
2438396 2440148
2448080 2449715
2462846 2462847
From the log-file:
Code: Select all
---------------------
Initial Commercial List
---------------------
0) 45067 46835 0:01:10.71
1) 52279 54031 0:01:10.07
2) 61963 63598 0:01:05.39
No change
Somehow, 2386117 gets added to each frame number in the txt file?
If I re-run manually, there is still the same offset.
Most of the time, Comskip works very well.
Using
comskip81_056_donators with default settings together with Comskip Monitored on MediaPortal 1.5.0.
Any ideas..?
Edit:
Ok, I tried to post the complete log file in a separate post, but it was too big to be accepted.
I zipped the files instead:
Code: Select all
http://www.hagastrom.se/comskipexample.zip
It is the show called _1 that has failed. The other one worked ok. These were two example recordings I made to gather test data. And no, I don't normally watch "Hem till gÄrden"...
Re: Frames out of bounds in txt but ok in log
Posted: Tue Nov 05, 2013 7:26 pm
by erik
Would you be able to upload the failing recording to my FTP server?
Re: Frames out of bounds in txt but ok in log
Posted: Tue Nov 05, 2013 11:30 pm
by IrmaIsfot
I have three recordings that gives this strange result, with frame counts above 2 million.
PM me about ftp details, and I will upload the smallest one; about 3 GB. I only have 5 mbit upstream, so it will take a while...
I also checked the results in older recordings:
During 2009 and 2010, the numbers in the txt file and the "Initial Commercial List" section in the log are always the same. These were made with an older version of Comskip, obviously.
More recent recordings have an added offset of between 4 and 14 frames in the txt file, compared to the values in the log file. The offset has been the same for all the "skips" in one recording, but seem to differ between recordings?
Since I discovered this "2 million frames phenomena" the other day, I've recorded several programs from that HD channel.
Most of these result in a txt file that seems ok, ie frame numbers within the file. However, comparing the txt file and the values in the log show strange results:
Code: Select all
FILE PROCESSING COMPLETE 135189 FRAMES AT 2500
-------------------
3260 8594
24837 32285
49248 56514
90141 97967
123014 130388
136341 136478
---------------------
Initial Commercial List
---------------------
0) 2173 7502 0:03:33.36
1) 23725 31159 0:04:57.91
2) 48074 55319 0:04:50.64
3) 88905 96712 0:05:13.04
4) 121746 129099 0:04:54.95
5) 135052 135190 0:00:05.48
1087
1112
1174
1236
1268
1289
The offset increases for each skip, to a point where the last skip point in the txt file is beyond the last frame: 136341, file length 135189 frames.
Another example, where the txt file results are beyond the file length:
Code: Select all
FILE PROCESSING COMPLETE 25057 FRAMES AT 2500
-------------------
50780 52603
61439 61473
---------------------
Initial Commercial List
---------------------
0) 14364 16187 0:01:12.91
1) 25023 25058 0:00:01.35
What additional processing is made after the log file is finished, and the txt file is written..?
Re: Frames out of bounds in txt but ok in log
Posted: Wed Nov 06, 2013 7:16 am
by IrmaIsfot
So I re-processed a couple of those recordings. It turns out the initial txt file was most correct, after all!
Code: Select all
FILE PROCESSING COMPLETE 136466 FRAMES AT 2500
-------------------
3260 8594
24837 32285
49248 56512
90143 97969
123016 130390
136343 136480
---------------------
Initial Commercial List
---------------------
0) 3246 8580 0:03:33.36
1) 24823 32271 0:04:57.91
2) 49234 56498 0:04:50.55
3) 90129 97955 0:05:13.03
4) 123002 130376 0:04:54.95
5) 136329 136467 0:00:05.47
The initial run was made while recording: "Launch at recording start". Maybe that is not reliable enough..?
The recordings with 2 million frames are the same, though.
Re: Frames out of bounds in txt but ok in log
Posted: Wed Nov 06, 2013 9:22 am
by erik
Could you set in your comskip.ini
output_timing=1
Rerun comskip on the 2 million frames and send me the generated ....timing.csv file?
Re: Frames out of bounds in txt but ok in log
Posted: Thu Dec 12, 2013 4:40 am
by peppzilla
I'm having the same problem, although not to this extent. Mine are shifted a handful of frames - just enough to be annoying. I was testing it by cutting a file into smaller pieces using VideoRedo. In every case the shorter recording resulted in a correct txt file with no offset. So I just fed the entire recording through VideoRedo and the resulting txt file had no offset. I don't know what VideoRedo does (the file size isn't exactly the same, so there is some modification), but whatever it is causes Comskip to not insert the offset. Looking at the ...timing.csv for both files shows one starting at what looks like the offset frame, while the other starts at 1.
One that works:
Code: Select all
type dts pts clock delta offset
v clock 0 0 0 0 0
v free 1 1 1 0 0
v free 2 2 2 0 0
v free 3 3 3 0 0
v free 4 4 4 0 0
One that doesn't:
Code: Select all
type dts pts clock delta offset
v init 11 11 11 0 0
v free 12 12 12 0 0
v free 13 13 13 0 0
v free 14 14 14 0 0
v free 15 15 15 0 0
Re: Frames out of bounds in txt but ok in log
Posted: Thu Dec 12, 2013 7:42 am
by erik
I found its a error in the decoding library I am using where the initial frame PTS is wrong.
I implemented some workaround and plan to release that soon.
Re: Frames out of bounds in txt but ok in log
Posted: Fri Dec 20, 2013 6:41 am
by peppzilla
That's great! This was driving me nuts. I thought there was a setting or something somewhere I was missing.
Re: Frames out of bounds in txt but ok in log
Posted: Sun Dec 22, 2013 11:18 am
by erik
Can you test build 0.81.57 released today?
Re: Frames out of bounds in txt but ok in log
Posted: Sun Dec 22, 2013 8:01 pm
by peppzilla
0.81.57 gives the same results as 0.81.56 (I noticed that the file sizes for comskip.exe were identical for the two builds, though).
I know that it used to work perfectly, so I ran the same files through 0.81.43. That build does indeed work perfectly, and the txt and log files are the same.