Re: mjpegtools 2.0.0 RC1

Hi,

Am 03.01.2011 19:08, schrieb Bernhard Praschinger:
> Hallo
>
>> I was mostly encoding animated stuff (the last thing I tried was Star
>> Trek: The Animated Series), which I guess is more difficult to get
>> right for rate control code.

Not only for rate control. Most compressions work by spending fewer bits
for the high frequency components of the images. Natural images are always
a bit blurry, so artifacts are hardly noticable. But cartoon or computer
generated stuff (images with sharp corners) are the second worst to
compress. The worst is noise.

>> Also your test video might have been too
>> short, I'd usually get between one and five underflows per full CD.
>
> I did now encode a movie with a playtime of about 88786PAL Frames or 59
> minutes, the final mpeg file has a size of 618MB. It was from a concert
> the played in the TV. And I did not get underruns but one buffer overrun
> (mplex message: data will arrive too late sent)
>
> Can you recommend something I can record so I have some animation I can
> encode ? (preferable something receivable using Astra satellite like:
> ORF/RTL/ZDF/3Sat/...)

ORF is on Astra now? I remember when they had to reduce the transmitter power
at the German border because of licensing issues (and under pressure from
RTL blabla)

>>> BTW: Do you know why the patch for the reduction parameter -4 -2
>>> did not make it into the CVS (You sent it to the developer
>>> mailinglist at the 05.09.2010) ? The patch lookes ok for me.  If no
>>> one objects I will add it to the CVS.
>> The first patch I sent, which enabled "-4 0" and "-2 0" to work
>> without crashing, actually made signal-to-noise ratio (PSNR) slightly
>> worse on some test encodings (while considerably slowing down
>> encoding), so I myself sent a message suggesting that it should not
>> be applied and instead a short patch which simply exits with an error
>> on these parameter settings should be used. However by now I've
>> learned that PSNR is not really that good a measure of encoding
>> quality. So maybe some other tests would have to be conducted. If you
>> or somebody else on this group could suggest a good mathematical test
>> of encoding quality that detects even relatively small changes, I
>> could do some more tests when I'm home again next week.
> Sorry, I don't have a idea for a mathematical test.

I wrote a bit about PSNR and SSIM here:

http://hirntier.blogspot.com/2010/01/video-quality-characterization.html

If you do fine-tuning with a single codec at high quality levels,
PSNR is very sensitive and is normally used. For comparing different codecs,
SSIM is better. There are some other methods as well, but not trivial to
implement.

Burkhard

------------------------------------------------------------------------------
Learn how Oracle Real Application Clusters (RAC) One Node allows customers
to consolidate database storage, standardize their database environment, and, 
should the need arise, upgrade to a full multi-node Oracle RAC database 
without downtime or disruption
http://p.sf.net/sfu/oracle-sfdevnl