Hi!
Thought folks might be interested in a couple numbers comparing
Apple's "Compressor" (new mpeg2 encoder included with DVD Studio
Pro 2) and 'mpeg2enc'. The encoding's being done on the G4 Powerbook.
A 1GHz G4 averages about 4 frames/sec on the encoding using mpeg2enc
and 5.8 fps with Apple's encoder. So mpeg2enc compares favorably
in speed on an Altivec based system. Be nice to try one of the
new dual 2GHz G5 systems sometime ;)
I've also a question :)
First the question about something I don't quite understand. When
would P frames end up being larger than I frames (given a good source
file)? Not that it's a "bug" but it did seem strange that the P frames
would be larger than the I frames.
The source material is from a laserdisc via S-Video cable to a
Canopus ADVC100 so any 'noise' is from the original film rather than
broadcast TV or VHS tape.
The mpeg2enc command used was (the DV file was converted to Y4M in
advance so it could be used multiple times without the overhead of
conversion each time):
mpeg2enc -f 8 -b 8500 -E -10 -q 6 -2 1 -K default -o foo.m2v < rhrh.y4m
No filtering being done at all.
INFO: [mplex] No. Pictures : 12961
INFO: [mplex] No. Groups : 889
INFO: [mplex] No. I Frames : 889 avg. size 35774 bytes
INFO: [mplex] No. P Frames : 3465 avg. size 43393 bytes
INFO: [mplex] No. B Frames : 8608 avg. size 30432 bytes
INFO: [mplex] Average bit-rate : 8214800 bits/sec
INFO: [mplex] Peak bit-rate : 8196000 bits/sec
Not a large variance between the I, P and B frames is there?
Using the 'tmpgenc' quantization matrices with the command:
mpeg2enc -f 8 -b 8500 -E -10 -q 6 -2 1 -K tmpgenc -o foo.m2v < rhrh.y4m
INFO: [mplex] No. Pictures : 12961
INFO: [mplex] No. Groups : 889
INFO: [mplex] No. I Frames : 889 avg. size 49247 bytes
INFO: [mplex] No. P Frames : 3465 avg. size 34919 bytes
INFO: [mplex] No. B Frames : 8608 avg. size 21420 bytes
INFO: [mplex] Average bit-rate : 6458000 bits/sec
INFO: [mplex] Peak bit-rate : 8209200 bits/sec
And finally with the kvcd matrices:
INFO: [mplex] No. Pictures : 12961
INFO: [mplex] No. Groups : 889
INFO: [mplex] No. I Frames : 889 avg. size 54368 bytes
INFO: [mplex] No. P Frames : 3465 avg. size 27293 bytes
INFO: [mplex] No. B Frames : 8608 avg. size 15870 bytes
INFO: [mplex] Average bit-rate : 5170000 bits/sec
INFO: [mplex] Peak bit-rate : 8138400 bits/sec
mpeg2enc -f 8 -b 8500 -E -10 -q 6 -2 1 -K kvcd -o foo.m2v < rhrh.y4m
By way of comparison using Apple's 2-pass VBR encoder (8500 maxbitrate,
7000 as requested average) using the same DV input file this is what
mplex says:
INFO: [mplex] No. Pictures : 12960
INFO: [mplex] No. Groups : 865
INFO: [mplex] No. I Frames : 865 avg. size 75850 bytes
INFO: [mplex] No. P Frames : 4320 avg. size 51659 bytes
INFO: [mplex] No. B Frames : 7776 avg. size 11511 bytes
INFO: [mplex] Average bit-rate : 6998800 bits/sec
INFO: [mplex] Peak bit-rate : 9281600 bits/sec
Using Apple's "Compressor" in 1-pass mode with a bitrate of 8500
gives:
INFO: [mplex] No. Pictures : 12960
INFO: [mplex] No. Groups : 865
INFO: [mplex] No. I Frames : 865 avg. size 60652 bytes
INFO: [mplex] No. P Frames : 4320 avg. size 54216 bytes
INFO: [mplex] No. B Frames : 7776 avg. size 21644 bytes
INFO: [mplex] Average bit-rate : 8417600 bits/sec
INFO: [mplex] Peak bit-rate : 10794800 bits/sec
Fairly close agreement between the 1pass mode of "Compressor" and
mpeg2enc on the avarage bitrate - and Apple's encoder has the I frames
being larger than the P frames as expected.
Notice that the I frame sizes are very different between mpeg2enc at
35774 bytes and Apple's encoder at 60652 bytes.
Any ideas why the I frames would be so different at the same bitrate?
It's almost like something is limiting/truncating the I frame size
calculations in mpeg2enc but I have no idea what that could be.
Interesting, eh? ;)
Cheers,
Steven Schultz
-------------------------------------------------------
This SF.Net email sponsored by: ApacheCon 2003,
16-19 November in Las Vegas. Learn firsthand the latest
developments in Apache, PHP, Perl, XML, Java, MySQL,
WebDAV, and more! http://www.apachecon.com/
_______________________________________________
Mjpeg-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/mjpeg-users
