Comskip goes HD

[erik]

I have removed the SD video size limitation in Comskip
The maxium video size is now 2000x1200
An experimental HD version of comskip can be downloaded from http://mk.kaashoek.com/gbpvr/comskipHDbeta.zip
All SD processing is as it used to be so you can use it for both SD and HD
Audio processing on HD is not yet operational so the result may be less accurate
Input can only be a Program Stream, Transport Stream does not work yet
I appriciate your feedback

[-stattik-]

Wow that was fast! I tested it out and it detected 6 commercial breaks. I'll test the output file tonight to check for accuracy.

[erik]

If you can run comskip on a mpeg file you can also run it on the resulting .txt file. Comskip will open a preview window allowing you to check the results visually by comparing the detected commercial start and stop points with the actual mpeg video, and even edit the results if not entirely correct (if I did not break anything)

[velocityboy]

erik, I'm curious if you've done any benchmarking with comskip, or if there's any way to benchmark it in the first place. I'll be building a new pc for gbpvr soon and performance of athlonxp vs. athlon64 vs. celeron vs. p4 would be very interesting. I wonder if the added cache of the P4, say, would overshadow a clock advantage of a cheaper celeron.
Maybe you could make a standard 100mb mpeg clip to run comskip against?

[erik]

Benchamrking is not that difficult.
You take one clip and run it on various PC's


I have benchmarked the following CPU's

2.4GHz Pentium4 without HT
1.8GHz Mobile pentium
2GH AMD64

The mobile pentium and the 2GH AMD64 are about the same speed, the 2.4GHz Pentium is the slowest.
I did not test Celeron but you can be sure its performance will be far worce.

[-stattik-]

Ok, so I ran a few ATSC files though ComskipHD but it missed all the commercials and labled parts of the program as commercials. I'll have to sort through the logs file to see if anything can be tweaked. One thing I noticed is that the "timestamp" in the lower left corner of comskip started at 08:17:00 instead of 00:00:00. I uploaded the logs if you're interested.

[erik]

I had a look at the logs you uploaded.

If I set
min_show_segment_length=150
I get

FILE PROCESSING COMPLETE 99355 FRAMES AT 2997
-------------------
1 2711
11928 18269
28575 35022
53555 60511
72809 81030
93325 99355

Which, looking at the debugwindow seems right but is completely different from the frame numbers you indicated.
The commercials found are consistent with the logo pattern.
Does the logo indeed disappear during commercials?

[-stattik-]

I set

min_show_segment_length=150

And the commercial groups were better defined but the timing is still off. I got roughly the same numbers as you it it looked like the commercial breaks got more and more out of synch between what comskip reported and what the debugwindow listed. Perhaps it's framerate related? Here's the differences.

Segment 1 Accurate
Segment 2 +2252 frames beyond what the debug window shows.
Segment 3 +4439
Segment 4 +8442
Segment 5 +11130
Segment 6 Unable to tell. The Debugwindow goes to frame 99154 but theres another 7:42 left of the show.

The reported commercial breaks seem pretty accurate but the distance between the breaks gets farther and farther apart.

[erik]

There seems to be a constant offset factor
Because we are dealing with frames the framerate can not be important.

Are the commercials as found aligned in the debug window?
Do the commercials start and stop where the debug bar has placed them?
I suspect the mpeg decoder I use.
Anyone experience with libmpeg2 and HD?

Is there someone who is willing to upload a HD mpeg PS of at least 20000 frames with a commercial in the last 5000 frames?

[-stattik-]

Is there someone who is willing to upload a HD mpeg PS of at least 20000 frames with a commercial in the last 5000 frames?

It's on its way.