On Sat, 20 Sep 2003, Richard Ellis wrote:
> Yes, only 5%, but it's a 5% file size change, with only a 0.5% change
> to the value of -N. Or in other words, the -N value looks to
> possibly have a 10x effect on resultant file size.
>
That's not quite the way to look at it I believe. Is it linear?
Probably not would be a guess. I don't think it's valid to
equate the scale 0 to 2.0 with any direct percentage change in
the file size. If the scale for -N had been set at a max of
.1 would you expect a 100% change in filesize?
It's probably also data sensitive and I suspect nonlinear. If you
used 1.0 what I doubt very much the change in filesize is going to
be 50% or any nice multiple of the .1 case.
I'd have to re-run the tests I did. Ah, I have the one table
(should be in the archives) from when I was testing libavcodec's
DV decoder compared to libdv (and just after -N took an argument).
4 encodings were done using -N of 0, 0.5, 1.0 and 1.5 (and 2
different -q values, and using yuvdenoise in -f and -l modes).
Quite an exhaustive (and time consuming ;)) set of runs.
From the "-q 5" test (no yuvdenoise):
q5-N0.0.txt: INFO: [mplex] Average bit-rate : 8301600 bits/sec
q5-N0.5.txt: INFO: [mplex] Average bit-rate : 8251600 bits/sec
q5-N1.0.txt: INFO: [mplex] Average bit-rate : 8074000 bits/sec
q5-N1.5.txt: INFO: [mplex] Average bit-rate : 8002800 bits/sec
So in my case going from 0.0 to 0.5 only made a difference of
50000 bits/sec out of ~8300.
But with the -q 6 run:
q6-N0.0.txt: INFO: [mplex] Average bit-rate : 8253200 bits/sec
q6-N0.5.txt: INFO: [mplex] Average bit-rate : 8041200 bits/sec
q6-N1.0.txt: INFO: [mplex] Average bit-rate : 7274000 bits/sec
q6-N1.5.txt: INFO: [mplex] Average bit-rate : 6437200 bits/sec
LOTS of variables involved - it's not as simple as changing one
parameter and trying to convert that into a simple relationship
with another parameter (i.e. changing one parameter which has
one range/scale and expecting that to relate to the file size).
I don't have that particular piece of DV data but I do have
one 7minute clip online that I can begin experimenting with - be
interesting to give the system a workout again.
Cheers,
Steven Schultz
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Mjpeg-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/mjpeg-users
