On Mon, 1 Nov 2004, Dik Takken wrote:
> > -q 4 is the practical limit with 5 being the usual value used.
>
> You mean that -q 4 can also trigger encoder bugs? I would like to get this
Ooops - I was unclear on that point. >= 4 is fine - by "practical
limit" I intended to say that "4 or greater has not been observed to
have problems". Values less than 4 (i.e. <=3) have been reported
to have errors.
You have mentioned seeing artifacts (induced by the errors in the
DCT/iDCT) at -q 1 and 2. There has been a report or two mentioning
the same problem with -q 3 (at least on IA32 systems). The Altivec
routines have code to check for overflow and then redo the calculations
to avoid the problem.
> one straight: Which q values can lead to encoding *errors*, not artifacts?
errors == artifacts in this case. The arithmetic overflow/truncation
is not only an error but it causes visible artifacts.
> I guess this method is safe, because it does not push mpeg2enc into
> encoding really close to the b value, which leads to artifacts and spikes.
It's not that the peak and average are too close (or the same). It's
that the quantizer scale ('qscale') at -q 1 thru 3 can cause
arithmetic errors (overflow) in the MMX/SSE routines.
There will always be spikes until a two pass (or back up and retry)
method of encoding is used.
> > think there is a mention about "values below 4 are extremes" in the
> > 'howto'.
>
> It said I should only use those values when I know what I'm doing. Well, I
:-)
> didn't because it didn't say what would happen. :) Now I know. :D It does
> tell about the danger of artifacting, but it does not mention that it can
> lead to encoding errors.
Hmmm, "errors" in encoding (often) show up as visible artifacts ;)
Artifacts are things present in the output that are not part of the
original input, in other word: error.
It's better to leave the qscale at a more conservative level and
process the data before it enters the encoding process. An
'unsharp' filter might be one thing to try (don't think there is
one at the moment - I have some notes based on the Gimp unsharp
filter but I haven't created any working code yet).
Cheers,
Steven Schultz
-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Mjpeg-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/mjpeg-users
